Skill instructions
atbash-manage2.83 KB
View saved version →
---
name: atbash-manage
description: Make an approved change to an existing Atbash agent from Codex. Use when the user wants to change an agent name, purpose, risk, or active state, or explicitly requests a new Atbash management authorization.
---
# Atbash Management
Every on-chain change requires a fresh browser authorization. A prior setup, active profile, existing browser login, or agent private key does not authorize a new owner-level mutation.
Start `manage start --host codex` through this skill's `scripts/atbash-control.mjs` launcher. Show the verification URL and code, then inspect the job after wallet verification. Build a non-secret plan JSON with one exact `update_agent` action using the discovered organization, network, and agent public key plus only the fields the user asked to change. Submit it with `manage plan <job-id> --input <path>` through the launcher.
The helper runs outside the Codex sandbox: inside it, even with network access granted, its DNS lookups fail. Run every launcher command, from the first one, with `sandbox_permissions` set to `"require_escalated"`, a one-sentence `justification` saying the Atbash management helper needs to reach Atbash outside the sandbox, and `prefix_rule` set to `["node", "<skill-directory>/scripts/atbash-control.mjs"]` with the same absolute launcher path as the command. If the shell tool has no `sandbox_permissions` parameter, Codex is already running without a sandbox: run the command as usual. If Codex rejects the escalation because its approval policy never asks, tell the user to switch Codex's permissions to Full access with `/permissions`. A helper error that says Atbash "could not be reached" means the command ran inside the sandbox, not that the service is down.
The user reviews and signs the exact change in Connect Atbash. Do not click approval controls or sign for them. Inspect and continue until the receipt is complete. If any approved step partially completed, report it and start a new management session for further work.
If the user wants to stop before signing, run `manage cancel <job-id>` through the launcher: it cancels the authorization, so its link stops working, and deletes the job's local files. Run it too when the session was cancelled, rejected, or expired; if the helper reports that the session had already ended, the cleanup still succeeded. Do not cancel after the user signed the change; let it finish and report the receipt.
Supported changes are agent `name`, `purpose`, `risk`, and `active`. Billing upgrades, ownership transfer, key rotation/recovery, certification, policies, mandate, and accountability require their dedicated dashboard flows until the control API explicitly advertises those actions.
Never read, print, transmit, or ask for an agent private key. Agent keys authorize ordinary Atbash activity only; they do not grant organization-owner permission.
Referenced files: 2 — file references are metadata, not package contents.
atbash-safety2.33 KB
View saved version →
---
name: atbash-safety
description: Check a specific proposed command or tool action against the selected local Atbash agent's safety policy, or inspect that agent's readiness and policy.
---
# Atbash Safety Check
Run explicit checks using the packaged helper in a local desktop thread with Node.js 22.13.0 or newer. Resolve `<skill-directory>` to this file's directory.
For readiness and policy, run:
```text
node <skill-directory>/scripts/atbash-safety.mjs status
```
For a specific action, send JSON on standard input to:
```text
node <skill-directory>/scripts/atbash-safety.mjs evaluate
```
The input is `{"toolName":"<actual tool>","args":{...},"context":"<brief task purpose>"}`. Use the exact proposed tool name and arguments. Context is optional and must be a short explanation of this action; do not send conversation history or transcripts. Clarify missing action details before checking. Do not execute a proposed command merely to evaluate it. Submit only the action information needed for the requested check; omit credentials and unrelated private data.
The helper uses the selected local `codex` profile and signs locally. It sends the tool name, SDK-redacted arguments, and action context to the package's Atbash service. No private key belongs in action input or output.
Read the JSON result even if the process exits nonzero:
- Only `allow: true` with `verdict: "ALLOW"` permits proceeding within the user's already-authorized task. A check by itself does not authorize executing the action.
- `HOLD`: stop the action and explain the returned review reason.
- `BLOCK`: stop the action and explain the returned policy reason.
- `ERROR`, missing or malformed output, timeout, or configuration failure: stop the action; report that a safe decision was unavailable. For profile setup, use `$atbash-setup`.
If the user asked only for an evaluation, report the result and do not execute the action. If the user also authorized execution, check the exact action immediately before executing it through its normal tool. A changed command or arguments need a new evaluation. Never use the helper as a command executor or treat an evaluation reference as reusable authorization.
State the checked action and relevant reason. This package does not automatically intercept other tools; never report agent readiness or a successful check as proof of universal protection.
Referenced files: 2 — file references are metadata, not package contents.
atbash-setup14.1 KB
View saved version →
---
name: atbash-setup
description: Set up, connect, verify, troubleshoot, or switch the Atbash Safety profile for Codex. Use when the user wants to sign up for Atbash, connect an existing Atbash agent, create a new agent, check setup progress, or repair local plugin configuration.
---
# Atbash Setup
Use the bundled control helper for onboarding. Keep private keys outside the conversation and outside tool arguments.
## Carry out setup
Treat a request to sign up, set up, or connect Atbash as a request to perform the workflow. Run the bundled commands yourself with the available execution tool; command examples below are for you to execute, not instructions to hand back to the user. Resolve `<skill-directory>` to the directory containing this SKILL.md, and use that installed copy's launcher throughout the job.
Handle session creation, progress checks, discovery, plan-file creation and submission, and local activation yourself. Ask only for missing user choices and actions requiring the user's identity or consent: wallet sign-in, review and signature of the exact proposal, local entry of an existing agent key, and enabling or trusting the host hook. Do not send the user to manually create an account, organization, or agent when the control API supports that step.
After a user completes verification or approval, inspect the same job and perform the next available step. Keep the job ID in the conversation so setup can resume without asking the user to run commands. Do not claim completion until the helper and status checks confirm it.
## Start or resume setup
The helper runs outside the Codex sandbox. It must resolve and reach the Atbash dashboard, the chain nodes and its own loopback key page, and write the Atbash configuration directory (`~/.config/atbash`, or `$ATBASH_CONFIG_DIR` when it is set). Inside the sandbox, even with network access granted, its DNS lookups fail, so do not try it sandboxed first and do not use `request_permissions` or `additional_permissions` for it.
Run every helper command, from the first one, with these shell tool parameters:
- `sandbox_permissions`: `"require_escalated"`
- `justification`: one sentence saying the Atbash setup helper needs to reach Atbash and save its configuration outside the sandbox.
- `prefix_rule`: `["node", "<skill-directory>/scripts/atbash-control.mjs"]`, the same absolute launcher path as the command, so the user can approve the helper once instead of on every step.
If the shell tool has no `sandbox_permissions` parameter, Codex is already running without a sandbox: run the command as usual. If Codex rejects the escalation because its approval policy never asks, tell the user to switch Codex's permissions to Full access with `/permissions`, then run the same command again. A helper error that says Atbash "could not be reached" or "could not write" means the command ran inside the sandbox, not that the service is down: run the same command again once with the parameters above.
Run the helper through this skill's `scripts/atbash-control.mjs` launcher, using the absolute skill directory:
```text
node "<skill-directory>/scripts/atbash-control.mjs" setup start --host codex
```
Until setup activates a profile, the hook allows only these setup steps, run outside the sandbox as above, and questions to the user, and denies everything else with "Atbash is not set up yet". Run each helper command as one plain command in exactly this form: no `cd`, `&&`, `;`, pipes, redirection, environment-variable prefixes, or command substitution, and no `--service` option. Commands in any other form are denied.
The result contains a public `verificationUri`, verification code, opaque job ID, and `planPath`. Show the URL and code to the user and ask them to complete wallet verification in **Connect Atbash**, then come back to the conversation. Never expose files under `~/.config/atbash/pending`, `credentials`, `profiles`, or `hosts`.
Inspect progress with `setup inspect <job-id>` through
Preview of the first 4,000 characters. Read full instructions →
Referenced files: 3 — file references are metadata, not package contents.