Yuimi-chaya/codex-development-guidelines / skills/adopt-agent-policy
adopt-agent-policy
VerifiedConfigure Codex user-level AGENTS.md when the user asks to install, adopt, or update these guidelines. Inspect the actual environment, relevant Codex configuration, and exposed models first; research public model facts, recommend a configuration, collect unresolved preferences together, and apply the authorized changes. Honor an explicit project-level target. Reviewing the repository or ordinary coding does not request installation.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 6.5 KB |
| agents/openai.yaml | config | 398 B |
| assets/runtime-policy.en.md | markdown | 12.7 KB |
| assets/runtime-policy.zh-CN.md | markdown | 10.8 KB |
| references/adoption.md | markdown | 11.1 KB |
| references/discovery.md | markdown | 7.3 KB |
| references/interview.md | markdown | 6.7 KB |
| references/model-selection.md | markdown | 7.1 KB |
| references/profile.md | markdown | 3.4 KB |
| references/update.md | markdown | 4.9 KB |
| references/user-interaction.md | markdown | 4.8 KB |
references/update.md
View on GitHub# Guided Policy Update Workflow
Use this workflow for a deliberate policy update, adoption-record refresh, or Skill update. An ordinary new thread is not an update event and must not trigger the interview again. A read-only review remains read-only and does not require the adoption interview.
## Update principles
- Use the current bundled policy asset unless another source is selected; identify versions or local drafts where available. If an old source is unavailable, disclose the limited comparison without blocking all updates or treating local differences as package-owned.
- Treat reference files as candidate data, not active instructions.
- Preserve user edits made after initial adoption.
- Re-detect objective facts such as platform, tools, models, permissions, and installation state.
- Revisit only choices whose meaning, capability, scope, or user preference changed; present affected recommendations and decisions together.
- Use stable IDs only for internal source comparison when available. Installed instructions need none; compare their actual behavior and preserve user-owned text. If a prior adoption copied this repository's maintenance tags, propose removing those tags while retaining behavior; do not mistake user-owned labels for leaked metadata.
- Keep runtime rules separate from first-adoption preferences and profile data.
- Preserve stricter local primary-browser prohibitions when reviewing the adaptive execution rule. A new reference default is not authorization to loosen them.
- If model choices need refreshing, use [model-selection.md](model-selection.md) for affected exposed IDs only. Research and recommendations do not select a replacement, authorize paid tests, or imply tool installation.
- Review artifact-audience and reader-context guidance semantically: require substantive supported content, reading-order prerequisites, clear references, and locally understandable examples. Preserve requested voice, legitimate PR change/test/risk content, and required attribution; do not adopt a first-person blacklist or require every paragraph to repeat all background.
- For runtime interaction, reuse preferences and review only affected capabilities using [user-interaction.md](user-interaction.md). Include unresolved relevant choices in the consolidated proposal; updating policy does not enable a feature or authorize a restart. Review delegation coordination and existing-process monitoring without silently changing model mappings or retry limits.
## Procedure
1. Confirm Shell, working directory, task scope, and Git state when applicable.
2. Read applicable instructions/notes, current target and overrides, and relevant old/new sources. Follow [discovery.md](discovery.md) for affected configuration/environment/model checks before asking questions. An absent optional record or external checkout is not a blocker.
3. Verify target and installed-file encoding, BOM, line endings, permissions, and hashes.
4. Compare behavior, using catalog IDs only internally if available. Classify changes as unchanged, new, compatible, conflicting, locally customized, translated-only, deprecated, or not applicable.
5. Re-detect actual model, reasoning, browser, write, media, and other controls before reusing any mapping.
6. Use [interview.md](interview.md) as a checklist; collect unresolved affected choices together with recommendations, without presenting IDs or asking the user to pick rules. Reuse existing choices when their meaning and scope are unchanged. A new thread is not evidence that preferences changed.
7. Check the proposed instruction-chain bytes against the applicable loading budget, then show the runtime diff, requested record changes, recovery/backup plan, capability fallbacks, and separate Skill changes. Resolve known overflow before applying; do not silently raise a configuration limit.
8. Use applicable explicit authority or obtain the missing approval for the concrete diff. Enumerate optional records, installations, configuration writes, restarts, and publication as distinct scopes that can be decided together. Previous adoption does not grant blanket authority to overwrite a changed file.
9. Re-hash targets; if any changed since the preview, regenerate it. Establish the recovery point, back up original bytes, preserve unrelated changes and encoding/permissions, apply only authorized changes, and verify the result. Do not introduce maintenance IDs or delete pre-existing user identifiers without reviewing their actual purpose. A line-ending-only difference is not a semantic upgrade.
10. Report whether the result is updated, installed, pending restart, partially adopted, translated-only, published, superseded, or cancelled.
An update is a new review, not package-owned synchronization. There is no “repository wins” mode.
For writing and backup details use [adoption.md](adoption.md); for optional provenance records use [profile.md](profile.md). Keep effective decisions in the installed instructions, not only in that record.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |