gstack Workflows

View saved version →

---
name: autoplan
description: Run coordinated product, design, engineering, and developer-experience plan reviews.
---

# Autoplan

Portable ChatGPT/Codex adaptation of the `autoplan` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `autoplan` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Inspect the request and relevant repository evidence before judging the plan.
2. State the objective, user outcome, constraints, and assumptions.
3. Challenge the proposal from the skill's named lens and identify missing decisions.
4. Compare meaningful alternatives when the choice is not obvious.
5. Produce a concrete recommendation with acceptance criteria, risks, and verification steps.
6. Stop before implementation unless the user also asked to build.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: benchmark
description: Measure performance and compare against a known baseline using available execution or browser tools.
---

# Benchmark

Portable ChatGPT/Codex adaptation of the `benchmark` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `benchmark` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Define the metric, workload, environment, and baseline before measuring.
2. Run the same bounded workload for each comparison target.
3. Record execution conditions and raw evidence.
4. Compare results without hiding variance or failed runs.
5. Do not claim benchmark numbers unless they were actually measured.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: benchmark-models
description: Compare model performance on the same bounded workflow with explicit scoring criteria.
---

# Benchmark Models

Portable ChatGPT/Codex adaptation of the `benchmark-models` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `benchmark-models` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Define the metric, workload, environment, and baseline before measuring.
2. Run the same bounded workload for each comparison target.
3. Record execution conditions and raw evidence.
4. Compare results without hiding variance or failed runs.
5. Do not claim benchmark numbers unless they were actually measured.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: browse
description: Drive the native gstack browser when available; otherwise use host browser capabilities without pretending gstack is running.
---

# Browse

Portable ChatGPT/Codex adaptation of the `browse` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `browse` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Detect a compatible local gstack browser runtime and shell before invoking native commands.
2. If present, follow the installed upstream gstack browser Skill and its safety controls.
3. If absent but the host has another browser capability, perform the user job with that capability and state that it is not the native gstack browser.
4. If neither exists, stop rather than fabricate browser execution.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

## Native bridge

On Codex, if a compatible original gstack runtime is installed and shell access exists, prefer its current `/browse` instructions. Otherwise use the host browser capability as the portable fallback.

Referenced files: 1

View saved version →

---
name: canary
description: Run post-deploy checks and surface regressions after a release.
---

# Canary

Portable ChatGPT/Codex adaptation of the `canary` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `canary` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Inspect current configuration and operational state.
2. Identify provider-specific facts from repository evidence rather than guessing.
3. Prefer reversible changes and preserve existing deployment conventions.
4. Execute checks only through available host tools.
5. Report what changed, what was observed, and what still needs a compatible runtime.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: careful
description: Apply an extra safety check before destructive or difficult-to-reverse operations.
---

# Careful

Portable ChatGPT/Codex adaptation of the `careful` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `careful` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Establish the exact allowed scope and operation before mutation.
2. Detect destructive, irreversible, credential, production, database, history-rewrite, or broad-delete actions.
3. Require explicit confirmation for materially risky actions.
4. Keep edits inside the agreed scope and stop on scope ambiguity.
5. Verify the resulting state when the host provides the needed tools.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: codex
description: Provide a second-opinion code or plan review using available Codex reasoning and workspace evidence.
---

# Codex

Portable ChatGPT/Codex adaptation of the `codex` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `codex` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Read the change, diff, plan, or code in scope.
2. Trace affected paths and identify behavior changes and blast radius.
3. Validate suspected findings against code, tests, contracts, and runtime evidence when available.
4. Report only actionable findings with severity, location, evidence, and consequence.
5. Separate confirmed defects from open questions.
6. Do not modify reviewed code unless the user separately asks for fixes.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

## Boundary

This skill is a review role, not an external model dependency. Use the current Codex/ChatGPT host reasoning and available workspace evidence. Do not claim that a second remote model was called unless a real tool did so.

Referenced files: 1

View saved version →

---
name: context-restore
description: Restore saved project context and verify it against the current repository state.
---

# Context Restore

Portable ChatGPT/Codex adaptation of the `context-restore` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `context-restore` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Inspect current repository state before reading or writing remembered project context.
2. Keep saved context factual: decisions, branch/commit state, verified findings, and next tasks.
3. Do not store secrets or unrelated personal data.
4. On restore, compare saved facts against current repository state and flag drift.
5. Write only when the user requested persistence and the host provides a workspace write capability.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: context-save
description: Save concise project context, decisions, git state, and remaining work into the workspace when writing is available.
---

# Context Save

Portable ChatGPT/Codex adaptation of the `context-save` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `context-save` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Inspect current repository state before reading or writing remembered project context.
2. Keep saved context factual: decisions, branch/commit state, verified findings, and next tasks.
3. Do not store secrets or unrelated personal data.
4. On restore, compare saved facts against current repository state and flag drift.
5. Write only when the user requested persistence and the host provides a workspace write capability.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: cso
description: Perform an evidence-backed security review using OWASP and STRIDE-oriented checks.
---

# CSO

Portable ChatGPT/Codex adaptation of the `cso` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `cso` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Establish assets, trust boundaries, inputs, privileged actions, and likely attackers.
2. Inspect the in-scope code and configuration for concrete weaknesses.
3. Use OWASP and STRIDE as coverage aids, not as a checklist substitute for evidence.
4. Validate exploitability and impact before assigning severity.
5. Recommend specific remediations and verification steps.
6. Do not change code unless the user separately requests remediation.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: design-consultation
description: Create or refine a design system and its implementation guidance.
---

# Design Consultation

Portable ChatGPT/Codex adaptation of the `design-consultation` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `design-consultation` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Inspect existing product, brand, components, and implementation constraints.
2. Define the design job and the criteria that will decide between directions.
3. Produce materially different options rather than cosmetic variations.
4. Compare them against usability, consistency, accessibility, feasibility, and the stated goal.
5. Select a direction and turn it into concrete implementation guidance.
6. Do not claim visual or browser verification unless the host actually performed it.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: design-html
description: Create production-oriented HTML and CSS from an approved design direction.
---

# Design HTML

Portable ChatGPT/Codex adaptation of the `design-html` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `design-html` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Inspect repository instructions and the existing implementation pattern.
2. Confirm the requested behavior and scope.
3. Make the smallest coherent implementation that fits existing conventions.
4. Preserve unrelated work and avoid broad rewrites without need.
5. Run the closest relevant checks available on the host.
6. Report changed files, verification evidence, and remaining limitations.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: design-review
description: Audit an implemented interface against design quality, usability, and consistency criteria.
---

# Design Review

Portable ChatGPT/Codex adaptation of the `design-review` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `design-review` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Inspect the implemented interface and the intended design reference or product goal.
2. Review hierarchy, spacing, typography, states, responsiveness, accessibility, and interaction behavior.
3. Distinguish objective defects from taste preferences.
4. Prioritize issues by user impact.
5. Fix only when authorized and re-check the affected states with actual host tools.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: design-shotgun
description: Generate and compare several materially different design directions before selecting one.
---

# Design Shotgun

Portable ChatGPT/Codex adaptation of the `design-shotgun` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `design-shotgun` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Inspect existing product, brand, components, and implementation constraints.
2. Define the design job and the criteria that will decide between directions.
3. Produce materially different options rather than cosmetic variations.
4. Compare them against usability, consistency, accessibility, feasibility, and the stated goal.
5. Select a direction and turn it into concrete implementation guidance.
6. Do not claim visual or browser verification unless the host actually performed it.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: devex-review
description: Audit a real developer workflow and measure friction against the actual path.
---

# DevEx Review

Portable ChatGPT/Codex adaptation of the `devex-review` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `devex-review` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Read the change, diff, plan, or code in scope.
2. Trace affected paths and identify behavior changes and blast radius.
3. Validate suspected findings against code, tests, contracts, and runtime evidence when available.
4. Report only actionable findings with severity, location, evidence, and consequence.
5. Separate confirmed defects from open questions.
6. Do not modify reviewed code unless the user separately asks for fixes.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: diagram
description: Create a technical diagram from a textual description and return editable source when the host supports artifacts.
---

# Diagram

Portable ChatGPT/Codex adaptation of the `diagram` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `diagram` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Confirm the source content and required output format.
2. Use the host artifact or Python capability when available.
3. Generate the actual artifact rather than only describing how to make it.
4. Verify that the output opens/parses and return the resulting file reference.
5. If the required artifact tool is unavailable, state that limitation clearly.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: document-generate
description: Generate practical documentation from code and verified behavior.
---

# Document Generate

Portable ChatGPT/Codex adaptation of the `document-generate` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `document-generate` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Inspect source code, tests, configuration, and existing docs for the behavior being documented.
2. Identify the target reader and documentation job.
3. Write only behavior supported by current evidence.
4. Keep examples executable and consistent with the repository.
5. Verify links, commands, and referenced paths when host tools permit.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: document-release
description: Update release-facing documentation to match shipped behavior.
---

# Document Release

Portable ChatGPT/Codex adaptation of the `document-release` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `document-release` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Inspect source code, tests, configuration, and existing docs for the behavior being documented.
2. Identify the target reader and documentation job.
3. Write only behavior supported by current evidence.
4. Keep examples executable and consistent with the repository.
5. Verify links, commands, and referenced paths when host tools permit.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: freeze
description: Restrict requested edits to an explicitly named directory or scope.
---

# Freeze

Portable ChatGPT/Codex adaptation of the `freeze` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `freeze` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Establish the exact allowed scope and operation before mutation.
2. Detect destructive, irreversible, credential, production, database, history-rewrite, or broad-delete actions.
3. Require explicit confirmation for materially risky actions.
4. Keep edits inside the agreed scope and stop on scope ambiguity.
5. Verify the resulting state when the host provides the needed tools.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: gstack
description: Route software product, planning, review, QA, debugging, design, security, release, documentation, browser, iOS, and safety requests to the right gstack workflow.
---

# gstack Router

Route a request to the smallest matching gstack Skill. If the user names a Skill directly, use it. For a complex delivery request, chain only the phases needed and keep planning, mutation, verification, and release gates distinct.

## Routing rules

1. Start with intent, not keywords.
2. Prefer one specialist Skill for a bounded request.
3. For a build request with material ambiguity or cross-module impact, use `spec` or the relevant plan review before implementation.
4. For defects, use `investigate` before fixing when root cause is unclear.
5. Use `review` before landing material changes.
6. Use `qa-only` when the user wants findings without edits, and `qa` when fixes are explicitly in scope.
7. Use safety Skills as constraints around another workflow, not as substitutes for it.
8. Native browser, pairing, gbrain, and iOS workflows require compatible local capabilities. Never simulate them.

## Available workflows

- `office-hours`: Reframe a product idea before implementation begins.
- `plan-ceo-review`: Challenge a plan from product and company-value angles before implementation.
- `plan-eng-review`: Review architecture, data flow, failure modes, edge cases, and test strategy before coding.
- `plan-design-review`: Review product and interface design dimensions before implementation.
- `plan-devex-review`: Review developer experience, time to first success, friction, and persona paths.
- `plan-tune`: Tune when the workflow should ask questions versus proceed with safe assumptions.
- `autoplan`: Run coordinated product, design, engineering, and developer-experience plan reviews.
- `design-consultation`: Create or refine a design system and its implementation guidance.
- `spec`: Turn vague intent into a precise executable specification with acceptance criteria.
- `review`: Review a change before landing and find defects that can pass CI but fail in production.
- `codex`: Provide a second-opinion code or plan review using available Codex reasoning and workspace evidence.
- `investigate`: Run systematic root-cause investigation before proposing a fix.
- `design-review`: Audit an implemented interface against design quality, usability, and consistency criteria.
- `design-shotgun`: Generate and compare several materially different design directions before selecting one.
- `design-html`: Create production-oriented HTML and CSS from an approved design direction.
- `devex-review`: Audit a real developer workflow and measure friction against the actual path.
- `qa`: Run end-to-end QA, fix authorized defects, and re-verify them when host tools permit.
- `qa-only`: Run end-to-end QA and report findings without changing code.
- `scrape`: Extract structured data from a web page using the safest available browser or web capability.
- `skillify`: Convert a proven repeatable workflow into a reusable Skill with clear triggers and checks.
- `ship`: Prepare a change for delivery by checking tests, review evidence, repository state, and release steps.
- `land-and-deploy`: Land an approved change, observe CI and deployment, and verify production health when host access permits.
- `canary`: Run post-deploy checks and surface regressions after a release.
- `landing-report`: Summarize delivery status and release queue state without modifying anything.
- `document-release`: Update release-facing documentation to match shipped behavior.
- `document-generate`: Generate practical documentation from code and verified behavior.
- `setup-deploy`: Inspect a repository and establish deployment configuration guidance without inventing provider details.
- `gstack-upgrade`: Check and update an installed native gstack runtime only when the host can execute local commands and the user authorizes it.
- `context-save`: Save concise project context, decisions, git state, and remaining work into the workspace when writing is available.
- `context-restore`: Restore saved project context and verify it against the current repository state.
- `learn`: Capture, inspect, and maintain project learnings backed by observed evidence.
- `retro`: Produce a retrospective from repository evidence, delivery outcomes, and recorded learnings.
- `health`: Assess codebase health using available type checks, linting, tests, dead-code signals, and repository evidence.
- `benchmark`: Measure performance and compare against a known baseline using available execution or browser tools.
- `benchmark-models`: Compare model performance on the same bounded workflow with explicit scoring criteria.
- `cso`: Perform an evidence-backed security review using OWASP and STRIDE-oriented checks.
- `setup-gbrain`: Configure gbrain integration only when the native local runtime and required credentials are available.
- `sync-gbrain`: Refresh gbrain project context only when the native local runtime is available and the user authorizes the sync.
- `browse`: Drive the native gstack browser when available; otherwise use host browser capabilities without pretending gstack is running.
- `open-gstack-browser`: Open the native visible gstack browser only on a compatible local host.
- `setup-browser-cookies`: Import browser cookies into native gstack only with explicit user authorization on a compatible local host.
- `pair-agent`: Pair an authorized agent with native gstack browser access while preserving scope and revocation controls.
- `ios-qa`: Run QA on a real iOS device only when a compatible local device bridge is available.
- `ios-fix`: Investigate and fix iOS defects with regression checks on a compatible local device workflow.
- `ios-design-review`: Review an iOS interface against Apple platform conventions using real device evidence when available.
- `ios-clean`: Remove development-only iOS bridge wiring before release after verifying the target scope.
- `ios-sync`: Regenerate native iOS debug bridge artifacts only on a compatible local host.
- `careful`: Apply an extra safety check before destructive or difficult-to-reverse operations.
- `freeze`: Restrict requested edits to an explicitly named directory or scope.
- `guard`: Combine destructive-operation checks with a strict edit scope.
- `unfreeze`: Remove a previously established edit-scope restriction when the user explicitly requests it.
- `make-pdf`: Turn markdown or structured content into a PDF using the host document or Python capabilities when available.
- `diagram`: Create a technical diagram from a textual description and return editable source when the host supports artifacts.

## Host portability

The original gstack project contains local runtime features. This plugin keeps its public workflow names and intent but maps operations to ChatGPT/Codex host capabilities. When a compatible original gstack install exists on Codex, specialist Skills may defer to the current installed upstream workflow for native execution.

Referenced files: 1

View saved version →

---
name: gstack-native-runtime
description: Bridge portable gstack workflows to an existing native gstack installation on a compatible Codex host without making it a hard plugin dependency.
---

# gstack Native Runtime Bridge

Use only when a workflow needs a capability implemented by the original local gstack runtime. First detect a compatible installation and shell. Read the installed current upstream Skill instead of assuming paths or behavior from an old version.

The native runtime is optional. Do not install, update, import credentials, pair agents, or change configuration unless the user requested that mutation. If no native runtime exists, return to the portable specialist workflow and use host capabilities where equivalent.

Referenced files: 1

View saved version →

---
name: gstack-upgrade
description: Check and update an installed native gstack runtime only when the host can execute local commands and the user authorizes it.
---

# gstack Upgrade

Portable ChatGPT/Codex adaptation of the `gstack-upgrade` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `gstack-upgrade` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Detect whether a compatible native gstack installation and local shell are actually available.
2. If available, inspect its current version/configuration before changing it.
3. Require explicit user authorization for installation, update, credential, or sync mutations.
4. Follow the upstream runtime workflow where available.
5. If unavailable, provide the portable alternative or stop with a clear capability limitation.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

## Surface note

This workflow is published for Codex discovery because its full behavior depends on a local runtime, device, browser session, or credential-aware command environment. ChatGPT can still discuss the workflow, but must not claim native execution.

Referenced files: 1

View saved version →

---
name: guard
description: Combine destructive-operation checks with a strict edit scope.
---

# Guard

Portable ChatGPT/Codex adaptation of the `guard` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `guard` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Establish the exact allowed scope and operation before mutation.
2. Detect destructive, irreversible, credential, production, database, history-rewrite, or broad-delete actions.
3. Require explicit confirmation for materially risky actions.
4. Keep edits inside the agreed scope and stop on scope ambiguity.
5. Verify the resulting state when the host provides the needed tools.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: health
description: Assess codebase health using available type checks, linting, tests, dead-code signals, and repository evidence.
---

# Health

Portable ChatGPT/Codex adaptation of the `health` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `health` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Read the change, diff, plan, or code in scope.
2. Trace affected paths and identify behavior changes and blast radius.
3. Validate suspected findings against code, tests, contracts, and runtime evidence when available.
4. Report only actionable findings with severity, location, evidence, and consequence.
5. Separate confirmed defects from open questions.
6. Do not modify reviewed code unless the user separately asks for fixes.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: host-workspace-operator
description: Safely inspect, search, modify, and verify files or repositories using the narrowest workspace capability available in ChatGPT or Codex.
---

# Host Workspace Operator

Use host-provided capabilities, not imaginary plugin-owned filesystem tools. Prefer read, list, search, grep, patch, write, shell, then Python according to the job. Inspect before mutation. Patch focused areas before broad rewrites. Respect repository instructions and user authorization. Never claim an operation ran unless the host produced evidence.

For mutations, report changed paths and run the closest relevant verification available. If the required capability does not exist, state that limitation rather than inventing a tool.

Referenced files: 1

View saved version →

---
name: investigate
description: Run systematic root-cause investigation before proposing a fix.
---

# Investigate

Portable ChatGPT/Codex adaptation of the `investigate` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `investigate` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Reproduce or establish the failure from logs, tests, code paths, or user evidence.
2. Form competing hypotheses instead of jumping to a fix.
3. Trace the relevant data/control path and eliminate hypotheses with evidence.
4. Identify the root cause and the smallest safe correction.
5. Apply a fix only when mutation is authorized.
6. Re-run the reproducer and relevant regression checks before calling it fixed.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: ios-clean
description: Remove development-only iOS bridge wiring before release after verifying the target scope.
---

# iOS Clean

Portable ChatGPT/Codex adaptation of the `ios-clean` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `ios-clean` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Detect whether the host has the required local iOS tooling, project, and device/simulator access.
2. Inspect the exact target and current bridge/configuration before changing anything.
3. Use real device or simulator evidence for QA claims.
4. Keep modifications scoped and reversible.
5. If compatible tooling is unavailable, provide analysis or setup guidance only and label execution as unavailable.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

## Surface note

This workflow is published for Codex discovery because its full behavior depends on a local runtime, device, browser session, or credential-aware command environment. ChatGPT can still discuss the workflow, but must not claim native execution.

Referenced files: 1

View saved version →

---
name: ios-design-review
description: Review an iOS interface against Apple platform conventions using real device evidence when available.
---

# iOS Design Review

Portable ChatGPT/Codex adaptation of the `ios-design-review` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `ios-design-review` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Detect whether the host has the required local iOS tooling, project, and device/simulator access.
2. Inspect the exact target and current bridge/configuration before changing anything.
3. Use real device or simulator evidence for QA claims.
4. Keep modifications scoped and reversible.
5. If compatible tooling is unavailable, provide analysis or setup guidance only and label execution as unavailable.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

## Surface note

This workflow is published for Codex discovery because its full behavior depends on a local runtime, device, browser session, or credential-aware command environment. ChatGPT can still discuss the workflow, but must not claim native execution.

Referenced files: 1

View saved version →

---
name: ios-fix
description: Investigate and fix iOS defects with regression checks on a compatible local device workflow.
---

# iOS Fix

Portable ChatGPT/Codex adaptation of the `ios-fix` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `ios-fix` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Detect whether the host has the required local iOS tooling, project, and device/simulator access.
2. Inspect the exact target and current bridge/configuration before changing anything.
3. Use real device or simulator evidence for QA claims.
4. Keep modifications scoped and reversible.
5. If compatible tooling is unavailable, provide analysis or setup guidance only and label execution as unavailable.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

## Surface note

This workflow is published for Codex discovery because its full behavior depends on a local runtime, device, browser session, or credential-aware command environment. ChatGPT can still discuss the workflow, but must not claim native execution.

Referenced files: 1

View saved version →

---
name: ios-qa
description: Run QA on a real iOS device only when a compatible local device bridge is available.
---

# iOS QA

Portable ChatGPT/Codex adaptation of the `ios-qa` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `ios-qa` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Detect whether the host has the required local iOS tooling, project, and device/simulator access.
2. Inspect the exact target and current bridge/configuration before changing anything.
3. Use real device or simulator evidence for QA claims.
4. Keep modifications scoped and reversible.
5. If compatible tooling is unavailable, provide analysis or setup guidance only and label execution as unavailable.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

## Surface note

This workflow is published for Codex discovery because its full behavior depends on a local runtime, device, browser session, or credential-aware command environment. ChatGPT can still discuss the workflow, but must not claim native execution.

Referenced files: 1

View saved version →

---
name: ios-sync
description: Regenerate native iOS debug bridge artifacts only on a compatible local host.
---

# iOS Sync

Portable ChatGPT/Codex adaptation of the `ios-sync` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `ios-sync` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Detect whether the host has the required local iOS tooling, project, and device/simulator access.
2. Inspect the exact target and current bridge/configuration before changing anything.
3. Use real device or simulator evidence for QA claims.
4. Keep modifications scoped and reversible.
5. If compatible tooling is unavailable, provide analysis or setup guidance only and label execution as unavailable.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

## Surface note

This workflow is published for Codex discovery because its full behavior depends on a local runtime, device, browser session, or credential-aware command environment. ChatGPT can still discuss the workflow, but must not claim native execution.

Referenced files: 1

View saved version →

---
name: land-and-deploy
description: Land an approved change, observe CI and deployment, and verify production health when host access permits.
---

# Land And Deploy

Portable ChatGPT/Codex adaptation of the `land-and-deploy` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `land-and-deploy` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Inspect branch state, diff, tests, review status, and release configuration.
2. Block on unresolved high-severity defects or missing required checks.
3. Treat push, PR creation, merge, deployment, and external mutations as explicit actions that require suitable authorization.
4. Execute only the actions the host can actually perform.
5. Verify the resulting commit/PR/deploy state and report exact evidence.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: landing-report
description: Summarize delivery status and release queue state without modifying anything.
---

# Landing Report

Portable ChatGPT/Codex adaptation of the `landing-report` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `landing-report` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Inspect the relevant repository and workflow state.
2. Summarize status, blockers, risk, and next action.
3. Do not mutate files, branches, issues, or external systems.
4. State any unavailable evidence explicitly.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: learn
description: Capture, inspect, and maintain project learnings backed by observed evidence.
---

# Learn

Portable ChatGPT/Codex adaptation of the `learn` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `learn` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Inspect current repository state before reading or writing remembered project context.
2. Keep saved context factual: decisions, branch/commit state, verified findings, and next tasks.
3. Do not store secrets or unrelated personal data.
4. On restore, compare saved facts against current repository state and flag drift.
5. Write only when the user requested persistence and the host provides a workspace write capability.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: make-pdf
description: Turn markdown or structured content into a PDF using the host document or Python capabilities when available.
---

# Make PDF

Portable ChatGPT/Codex adaptation of the `make-pdf` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `make-pdf` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Confirm the source content and required output format.
2. Use the host artifact or Python capability when available.
3. Generate the actual artifact rather than only describing how to make it.
4. Verify that the output opens/parses and return the resulting file reference.
5. If the required artifact tool is unavailable, state that limitation clearly.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: office-hours
description: Reframe a product idea before implementation begins.
---

# Office Hours

Portable ChatGPT/Codex adaptation of the `office-hours` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `office-hours` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Inspect the request and relevant repository evidence before judging the plan.
2. State the objective, user outcome, constraints, and assumptions.
3. Challenge the proposal from the skill's named lens and identify missing decisions.
4. Compare meaningful alternatives when the choice is not obvious.
5. Produce a concrete recommendation with acceptance criteria, risks, and verification steps.
6. Stop before implementation unless the user also asked to build.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: open-gstack-browser
description: Open the native visible gstack browser only on a compatible local host.
---

# Open gstack Browser

Portable ChatGPT/Codex adaptation of the `open-gstack-browser` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `open-gstack-browser` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Detect a compatible local gstack browser runtime and shell before invoking native commands.
2. If present, follow the installed upstream gstack browser Skill and its safety controls.
3. If absent but the host has another browser capability, perform the user job with that capability and state that it is not the native gstack browser.
4. If neither exists, stop rather than fabricate browser execution.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

## Surface note

This workflow is published for Codex discovery because its full behavior depends on a local runtime, device, browser session, or credential-aware command environment. ChatGPT can still discuss the workflow, but must not claim native execution.

Referenced files: 1

View saved version →

---
name: pair-agent
description: Pair an authorized agent with native gstack browser access while preserving scope and revocation controls.
---

# Pair Agent

Portable ChatGPT/Codex adaptation of the `pair-agent` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `pair-agent` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Require a compatible local gstack browser runtime.
2. Treat cookies, authenticated tabs, tokens, pairing keys, and account sessions as sensitive.
3. Ask for explicit authorization before importing credentials, pairing agents, or changing grants.
4. Apply least privilege, scope limits, and revocation checks.
5. Do not print secrets or persist them in plugin files.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

## Surface note

This workflow is published for Codex discovery because its full behavior depends on a local runtime, device, browser session, or credential-aware command environment. ChatGPT can still discuss the workflow, but must not claim native execution.

Referenced files: 1

View saved version →

---
name: plan-ceo-review
description: Challenge a plan from product and company-value angles before implementation.
---

# Plan CEO Review

Portable ChatGPT/Codex adaptation of the `plan-ceo-review` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `plan-ceo-review` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Inspect the request and relevant repository evidence before judging the plan.
2. State the objective, user outcome, constraints, and assumptions.
3. Challenge the proposal from the skill's named lens and identify missing decisions.
4. Compare meaningful alternatives when the choice is not obvious.
5. Produce a concrete recommendation with acceptance criteria, risks, and verification steps.
6. Stop before implementation unless the user also asked to build.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: plan-design-review
description: Review product and interface design dimensions before implementation.
---

# Plan Design Review

Portable ChatGPT/Codex adaptation of the `plan-design-review` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `plan-design-review` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Inspect the request and relevant repository evidence before judging the plan.
2. State the objective, user outcome, constraints, and assumptions.
3. Challenge the proposal from the skill's named lens and identify missing decisions.
4. Compare meaningful alternatives when the choice is not obvious.
5. Produce a concrete recommendation with acceptance criteria, risks, and verification steps.
6. Stop before implementation unless the user also asked to build.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: plan-devex-review
description: Review developer experience, time to first success, friction, and persona paths.
---

# Plan DevEx Review

Portable ChatGPT/Codex adaptation of the `plan-devex-review` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `plan-devex-review` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Inspect the request and relevant repository evidence before judging the plan.
2. State the objective, user outcome, constraints, and assumptions.
3. Challenge the proposal from the skill's named lens and identify missing decisions.
4. Compare meaningful alternatives when the choice is not obvious.
5. Produce a concrete recommendation with acceptance criteria, risks, and verification steps.
6. Stop before implementation unless the user also asked to build.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: plan-eng-review
description: Review architecture, data flow, failure modes, edge cases, and test strategy before coding.
---

# Plan Eng Review

Portable ChatGPT/Codex adaptation of the `plan-eng-review` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `plan-eng-review` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Inspect the request and relevant repository evidence before judging the plan.
2. State the objective, user outcome, constraints, and assumptions.
3. Challenge the proposal from the skill's named lens and identify missing decisions.
4. Compare meaningful alternatives when the choice is not obvious.
5. Produce a concrete recommendation with acceptance criteria, risks, and verification steps.
6. Stop before implementation unless the user also asked to build.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: plan-tune
description: Tune when the workflow should ask questions versus proceed with safe assumptions.
---

# Plan Tune

Portable ChatGPT/Codex adaptation of the `plan-tune` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `plan-tune` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Inspect the request and relevant repository evidence before judging the plan.
2. State the objective, user outcome, constraints, and assumptions.
3. Challenge the proposal from the skill's named lens and identify missing decisions.
4. Compare meaningful alternatives when the choice is not obvious.
5. Produce a concrete recommendation with acceptance criteria, risks, and verification steps.
6. Stop before implementation unless the user also asked to build.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: qa
description: Run end-to-end QA, fix authorized defects, and re-verify them when host tools permit.
---

# QA

Portable ChatGPT/Codex adaptation of the `qa` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `qa` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Define the user journeys and expected outcomes in scope.
2. Use real browser, app, or execution tools when the host exposes them.
3. Record reproducible failures with steps, evidence, and severity.
4. If fixes are authorized, make focused changes and re-run the failing path.
5. Run a small regression pass around each fix.
6. Never claim a click, browser session, screenshot, or test run that did not occur.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: qa-only
description: Run end-to-end QA and report findings without changing code.
---

# QA Only

Portable ChatGPT/Codex adaptation of the `qa-only` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `qa-only` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Define user journeys and expected outcomes.
2. Exercise them with actual host tools when available.
3. Record reproducible findings with evidence, severity, and likely impact.
4. Do not edit code or configuration.
5. Clearly mark checks that could not be executed on the current host.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: retro
description: Produce a retrospective from repository evidence, delivery outcomes, and recorded learnings.
---

# Retro

Portable ChatGPT/Codex adaptation of the `retro` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `retro` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Gather the evidence relevant to the period or work being analyzed.
2. Separate observed outcomes from interpretation.
3. Identify recurring friction, successful patterns, and unresolved issues.
4. Produce specific follow-up actions tied to evidence.
5. Do not invent activity metrics that are not present in the source data.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: review
description: Review a change before landing and find defects that can pass CI but fail in production.
---

# Review

Portable ChatGPT/Codex adaptation of the `review` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `review` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Read the change, diff, plan, or code in scope.
2. Trace affected paths and identify behavior changes and blast radius.
3. Validate suspected findings against code, tests, contracts, and runtime evidence when available.
4. Report only actionable findings with severity, location, evidence, and consequence.
5. Separate confirmed defects from open questions.
6. Do not modify reviewed code unless the user separately asks for fixes.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: sandbox-python-executor
description: Use host-native Python for deterministic parsing, hashing, archive inspection, validation, transformations, and executable verification.
---

# Sandbox Python Executor

Use the host Python capability when deterministic execution materially improves correctness. Suitable jobs include manifest parsing, package inspection, hashing, structured validation, and file transformation. Do not use Python merely to imitate a safer file-read operation.

Actually execute the code before reporting outputs. Do not expose secrets. Treat target repository scripts as untrusted until inspected. If Python is unavailable, keep execution-dependent conclusions unverified.

Referenced files: 1

View saved version →

---
name: scrape
description: Extract structured data from a web page using the safest available browser or web capability.
---

# Scrape

Portable ChatGPT/Codex adaptation of the `scrape` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `scrape` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Prefer a host browser or web capability when one is available.
2. Establish the exact page, fields, and output schema before extraction.
3. Minimize navigation and avoid actions that change external state unless explicitly requested.
4. Validate extracted data for completeness and obvious parsing errors.
5. If browser capability is absent, explain the limitation instead of simulating browser output.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: setup-browser-cookies
description: Import browser cookies into native gstack only with explicit user authorization on a compatible local host.
---

# Setup Browser Cookies

Portable ChatGPT/Codex adaptation of the `setup-browser-cookies` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `setup-browser-cookies` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Require a compatible local gstack browser runtime.
2. Treat cookies, authenticated tabs, tokens, pairing keys, and account sessions as sensitive.
3. Ask for explicit authorization before importing credentials, pairing agents, or changing grants.
4. Apply least privilege, scope limits, and revocation checks.
5. Do not print secrets or persist them in plugin files.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

## Surface note

This workflow is published for Codex discovery because its full behavior depends on a local runtime, device, browser session, or credential-aware command environment. ChatGPT can still discuss the workflow, but must not claim native execution.

Referenced files: 1

View saved version →

---
name: setup-deploy
description: Inspect a repository and establish deployment configuration guidance without inventing provider details.
---

# Setup Deploy

Portable ChatGPT/Codex adaptation of the `setup-deploy` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `setup-deploy` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Inspect current configuration and operational state.
2. Identify provider-specific facts from repository evidence rather than guessing.
3. Prefer reversible changes and preserve existing deployment conventions.
4. Execute checks only through available host tools.
5. Report what changed, what was observed, and what still needs a compatible runtime.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: setup-gbrain
description: Configure gbrain integration only when the native local runtime and required credentials are available.
---

# Setup Gbrain

Portable ChatGPT/Codex adaptation of the `setup-gbrain` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `setup-gbrain` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Detect whether a compatible native gstack installation and local shell are actually available.
2. If available, inspect its current version/configuration before changing it.
3. Require explicit user authorization for installation, update, credential, or sync mutations.
4. Follow the upstream runtime workflow where available.
5. If unavailable, provide the portable alternative or stop with a clear capability limitation.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

## Surface note

This workflow is published for Codex discovery because its full behavior depends on a local runtime, device, browser session, or credential-aware command environment. ChatGPT can still discuss the workflow, but must not claim native execution.

Referenced files: 1

View saved version →

---
name: ship
description: Prepare a change for delivery by checking tests, review evidence, repository state, and release steps.
---

# Ship

Portable ChatGPT/Codex adaptation of the `ship` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `ship` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Inspect branch state, diff, tests, review status, and release configuration.
2. Block on unresolved high-severity defects or missing required checks.
3. Treat push, PR creation, merge, deployment, and external mutations as explicit actions that require suitable authorization.
4. Execute only the actions the host can actually perform.
5. Verify the resulting commit/PR/deploy state and report exact evidence.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: skillify
description: Convert a proven repeatable workflow into a reusable Skill with clear triggers and checks.
---

# Skillify

Portable ChatGPT/Codex adaptation of the `skillify` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `skillify` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Start from a workflow that has been observed or clearly specified.
2. Define a narrow trigger, user job, inputs, outputs, stop conditions, and mutation boundary.
3. Map operations to host capabilities rather than hard-coded proprietary tool names.
4. Add verification and negative routing cases.
5. Package the result using the current Agent Skills contract and validate the files.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: spec
description: Turn vague intent into a precise executable specification with acceptance criteria.
---

# Spec

Portable ChatGPT/Codex adaptation of the `spec` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `spec` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Inspect the request and relevant repository evidence before judging the plan.
2. State the objective, user outcome, constraints, and assumptions.
3. Challenge the proposal from the skill's named lens and identify missing decisions.
4. Compare meaningful alternatives when the choice is not obvious.
5. Produce a concrete recommendation with acceptance criteria, risks, and verification steps.
6. Stop before implementation unless the user also asked to build.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1

View saved version →

---
name: sync-gbrain
description: Refresh gbrain project context only when the native local runtime is available and the user authorizes the sync.
---

# Sync Gbrain

Portable ChatGPT/Codex adaptation of the `sync-gbrain` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `sync-gbrain` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Detect whether a compatible native gstack installation and local shell are actually available.
2. If available, inspect its current version/configuration before changing it.
3. Require explicit user authorization for installation, update, credential, or sync mutations.
4. Follow the upstream runtime workflow where available.
5. If unavailable, provide the portable alternative or stop with a clear capability limitation.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

## Surface note

This workflow is published for Codex discovery because its full behavior depends on a local runtime, device, browser session, or credential-aware command environment. ChatGPT can still discuss the workflow, but must not claim native execution.

Referenced files: 1

View saved version →

---
name: unfreeze
description: Remove a previously established edit-scope restriction when the user explicitly requests it.
---

# Unfreeze

Portable ChatGPT/Codex adaptation of the `unfreeze` workflow from `garrytan/gstack`. Preserve the original job and safety intent while mapping execution to capabilities the current host actually exposes.

## When to use

Use this Skill when the user explicitly names `unfreeze` or asks for the same job described above.

## Host contract

- Inspect repository or file evidence before making claims about the current state.
- Use host-native read, list, search, grep, patch, write, shell, browser, computer, and Python capabilities only when they actually exist.
- Never claim commands, tests, browser actions, device actions, file writes, Git operations, or external mutations that were not executed.
- Prefer read-only discovery before mutation.
- Respect repository instructions and preserve unrelated work.
- When the original native gstack runtime is available in Codex, it may be used as an implementation detail after inspecting the installed upstream Skill. Do not hard-code Claude-only paths as a requirement.

## Workflow

1. Establish the exact allowed scope and operation before mutation.
2. Detect destructive, irreversible, credential, production, database, history-rewrite, or broad-delete actions.
3. Require explicit confirmation for materially risky actions.
4. Keep edits inside the agreed scope and stop on scope ambiguity.
5. Verify the resulting state when the host provides the needed tools.

## Completion

Return the decision, findings, changed artifacts if any, executed verification, skipped checks, and remaining blockers.

Referenced files: 1