ForgeMind

View saved version →

---
name: forgemind-commands
description: "Use when a ForgeMind user wants to discover or deliberately select an advanced internal workflow."
---

# ForgeMind Commands

Show this concise choice list, then ask for a selection only when the user has not already stated one:

1. **Leap** — autonomously build a disruptive, tested MVP.
2. **Spark** — generate radical AI product directions.
3. **Venture** — validate market, positioning, pricing, and business-case assumptions.
4. **Ship** — implement and verify a bounded release.
5. **Growth** — plan activation, retention, monetization, and value-proof experiments.
6. **Portfolio** — rank AI-native product bets for the repository.
7. **Twin** — model workflows, integrations, and knowledge gaps in the application.
8. **Evolve UI** — plan a reversible, outcome-first UX experiment.
9. **Autopilot** — run a policy-bounded, goal-driven delivery mission.
10. **GUI Draft umsetzen** — implement a user-selected local PNG design draft with measured visual verification.

After the user selects one, invoke the matching advanced CLI command with their stated goal and report its evidence boundary, next action, hard stops, and artifact location. If no option clearly fits, use Compass instead.

For **GUI Draft umsetzen**, follow the Design Fidelity workflow exactly: create or load three local PNG proposals, let the user explicitly select one, persist a matching control contract, then run `design-fidelity prepare`. After `prepare` returns `implementation-ready`, you MUST inspect the selected draft and control contract, edit only matching allowed workspace UI files to implement that selected draft, run safe project verification, then run the exact measured verification command returned by `prepare`. Do not stop after `prepare`, and do not present the work order as an implementation. Keep changes only when the verification passes and the measured difference decreases.

Do not expose specialist routes as separate Marketplace skills. Do not claim market, customer, pricing, or delivery outcomes without returned evidence. Never run external, production, destructive, spending, or irreversible actions without the required configured adapter, policy, and approval.

Referenced files: 1

View saved version →

---
name: forgemind-compass
description: "Use when a product, app, or delivery goal is unclear and ForgeMind should choose the strongest safe starting path."
---

# ForgeMind One

Use this as the default start. Ask no question unless the answer changes safety, cost, production access, or a hard stop. Run `node <plugin-root>/bin/forgemind.mjs one --goal "<outcome>" --artifacts workspace --json`; it creates the canonical Foundation chain, starts the bounded Autopilot mission, and returns one next safe action. Run `node <plugin-root>/bin/forgemind.mjs status --artifacts workspace --json` to resume.

When invoked without user text, load `playbooks/zero-input-defaults.md` and derive a safe goal before running ForgeMind One. Marketplace installation does not create a global `forgemind` shell command; always use this bundled runner.

Compass selects and applies the appropriate internal route—discovery, transformation, market validation, delivery, or autonomous execution—without exposing a specialist-skill menu. For read-only application QA, use `$forgemind-xray` explicitly.

Referenced files: 1

View saved version →

---
name: forgemind-design-fidelity
description: Use when local PNG design references must be implemented and verified against a local web UI.
---

# ForgeMind Design Fidelity

## Canonical Product Design flow

Use this as the only flow for a Product Design handoff. When the user invokes Fidelity without an existing selected proposal, call `@Product Design` to create exactly three alternatives and stage their local PNG files in the workspace. Persist them:

`node <plugin-root>/bin/forgemind.mjs design-fidelity propose --inputs <proposal-1.png,proposal-2.png,proposal-3.png> --route <local-or-test-url> --viewport desktop --goal <design-goal> --artifacts workspace --json`

Present the three persisted local images with their proposal IDs, titles, and rationales. Ask exactly one decision question: which proposal should be implemented? Never infer a preference from ordering, filename, or recency. Persist the answer, including a short reason when the user gives one:

`node <plugin-root>/bin/forgemind.mjs design-fidelity select --proposal-set <set-id> --proposal <proposal-id> --reason <user-reason> --artifacts workspace --json`

Inspect only the immutable selected draft. Derive an observable control contract from it, recording uncertainty as an assumption rather than asking routine follow-up questions. Persist the contract before source changes:

`node <plugin-root>/bin/forgemind.mjs design-fidelity contract --contract '<json>' --artifacts workspace --json`

Create the implementation work order. `prepare` is deliberately not named `apply`: the CLI never pretends to have edited source files itself.

`node <plugin-root>/bin/forgemind.mjs design-fidelity prepare --proposal-set <set-id> --proposal <proposal-id> --control-contract <contract-id> --artifacts workspace --json`

Implement only the selected draft and work order. Do not reinterpret or redesign its observable layout, copy, controls, states, or safe interactions. Edit only matching workspace UI source, styles, templates, or local assets; never edit manifests, lockfiles, `.env` files, infrastructure, deployment, payment, identity, or production configuration.

After each source change, run the project's safe verification and then the exact verification command returned by `prepare` with recorded control observations. Keep an edit only if verification passes and the newly measured difference decreases. Fidelity records iterations, blocks a regression, and blocks further attempts once the contract limit is reached. Stop and report the exact evidence gap if the target is not local/test, a control is not evidenced, or the report is blocked.

Do not install packages, use external browsers, log in, submit forms, or claim a match without the current measured report.

Referenced files: 1

View saved version →

---
name: forgemind-guide
description: "Use when a ForgeMind user has an idea, existing project, or quality concern and wants a guided next step."
---

# Guided ForgeMind

Ask only for missing routing inputs, one at a time:

1. Starting context: `idea`, `project`, or `quality`.
2. Desired outcome: `improve`, `mvp`, or `ship`.
3. Working style: `guided` or `autonomous`.

Do not ask for a value the user supplied clearly. Run only:

`node <plugin-root>/bin/forgemind.mjs guide --context <context> --outcome <outcome> --mode <mode> --artifacts workspace --json`

It recommends an internal route only. State the recommendation, why it was selected, its alternative route when present, the next public action, and the returned safety boundary. Continue through Compass unless the recommendation is Xray; existing hard stops remain in force.

Referenced files: 1

View saved version →

---
name: forgemind-innovate
description: "Use when a SaaS team needs disruptive, AI-central USP hypotheses, rather than generic AI features."
---

# ForgeMind Innovate

Use this explicit journey for disruptive SaaS innovation: AI-central product bets, defensible moats, activation, retention, pricing, tenant safety, integration health, and staged release cohorts.

Ask only for a missing product outcome. Do not ask for a technical solution before examining the opportunity report. Then run:

`node <plugin-root>/bin/forgemind.mjs innovation saas --goal "<outcome>" --artifacts workspace --json`

Return the highest-ranked opportunity card, the interaction it replaces, its moat, first experiment, metric, guardrails, kill condition, and the recommended next action. Clearly distinguish imported evidence from repository-aware assumptions.

This command creates local planning evidence only. It does not contact customers, change billing, read production tenants, or invoke external integrations. Hold when tenant scope, permissions, or audit evidence are missing.

Referenced files: 1

View saved version →

---
name: forgemind-xray
description: Use when an existing application needs autonomous, read-only quality testing across GUI, API, CLI, and local integration surfaces.
---

# ForgeMind Xray

You MUST execute `node <plugin-root>/bin/forgemind.mjs xray run` before analysing, planning, scoring, or reporting. Marketplace installation does not create a global `forgemind` shell command.
Do not return a test plan, score, or report before the command has completed and its execution receipts have been inspected.

For an explicit local/test web URL, use the internal Codex Browser as Xray's preferred GUI executor. Start with a discovery pass and inspect `mission.criticalFlows`, `mission.testUrl`, and the returned prerequisites:

`node <plugin-root>/bin/forgemind.mjs xray run --goal "<scope>" --test-url <loopback-url> --adapters browser --artifacts workspace --json`

The URL MUST use `localhost`, `127.0.0.0/8`, `::1`, or a reserved `.test` host. Before Browser control, invoke the internal Browser skill and use its in-app Browser binding. For each safe route in `mission.criticalFlows`, execute every declared desktop and mobile viewport at the explicit same-origin local/test URL. Create or claim one tab, inspect its visible DOM, and execute only safe, non-destructive interactions: page loads, same-origin opted-in links, and non-submitting invalid-input validation.

The internal Codex Browser does not submit forms or perform login, download, upload, save, delete, account, administration, payment, deploy, publish, credential, or other consequential actions. Do not inspect cookies, local storage, profiles, passwords, or session stores. Do not start an app server, install packages, download a browser runtime, use another browser family, or invent a result.

For every attempted safe flow, capture observable before/after state and a screenshot artifact. Form a complete Browser receipt with `executor: "internal-browser"`, `viewport`, `surfaceId`, `control`, `status`, `componentIds`, `evidence`, `url`, `coverageArea`, `controlLabel`, `action`, `expected`, `actual`, and `reproduction`. The evidence must be workspace-local and refer to the current Xray run. Then execute:
Provide complete receipts only through `--gui-receipts`; incomplete observations remain Xray gaps.

`node <plugin-root>/bin/forgemind.mjs xray run --goal "<scope>" --test-url <loopback-url> --adapters browser --gui-receipts '<json-array>' --artifacts workspace --json`

Only that second CLI result is canonical: inspect its execution receipts, coverage, gaps, score, and improvement proposals before reporting. A visible Browser tab alone is not evidence.

If Browser control, the URL, evidence capture, or an eligible safe flow is unavailable, do not pass a successful GUI receipt. Preserve and report the CLI's returned Browser gap and next action, then continue command and API tests. If a flow is blocked or intentionally skipped, only pass a complete receipt with its truthful `blocked` or `skipped` status.

For direct CLI use without the internal Browser, Xray retains its workspace-local Playwright adapter. It runs only safe local checks and records unavailable controls as test gaps rather than claiming coverage. Playwright requires a declared local package and Chromium runtime; preserve `FM_XRAY_PLAYWRIGHT_UNAVAILABLE` with its exact next action when either is unavailable.

For local native GUI or mobile testing, Xray uses Android ADB only when it detects an Android surface and exactly one authorized emulator; it does not choose physical devices autonomously. Preserve `FM_XRAY_ANDROID_EMULATOR_UNAVAILABLE` or `FM_XRAY_ADB_UNAVAILABLE` when appropriate.

For visual regression testing, configure an explicit local web target plus `web.visualBaseline.enabled` and a `thresholdPercent` in the project-local ForgeMind Xray configuration. Review the UI first, then establish a baseline with `xray baseline`; normal `xray run` compares only workspace-local PNG screenshots against it. Configured navigation and API time budgets become findings only when receipt timing evidence exists. Native desktop and iOS simulator testing remain explicit environment gaps; never claim Android execution covers them.

Finally run `node <plugin-root>/bin/forgemind.mjs xray status --artifacts workspace --json` and hand off its detailed findings, explicit gaps, evidence references, Browser-covered areas, and informative score.

Referenced files: 1