Repository navigation
docs: update agent instructions for tstack - #8
Conversation
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: fa58447c58
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| - Frontmatter keys: `name` (equals the directory), `description`, and optionally `license`, `allowed-tools`, `disable-model-invocation`. Some harnesses ignore `disable-model-invocation`, so write descriptions that only match an explicit request. | ||
| - Cross-family model panels go through `skills/interrogate/references/bridge.md`. Never call another vendor's CLI outside it. | ||
| - Write for any harness. Say "spawn a subagent" and "ask the user", never a harness's tool name. Name model roles ("default", "fast", "strongest"), never model slugs. `BANNED_TERMS` in `check.mjs` enforces this. A file that must name a harness path, such as a per-harness table, gets an `ALLOWED_TERMS` entry. | ||
| - Frontmatter: `name` (equal to the directory, lowercase letters, digits, and single hyphens, max 64), `description` (max 1024), and optionally `license`, `compatibility`, `metadata`, `allowed-tools`, `disable-model-invocation`. |
There was a problem hiding this comment.
When a skill needs compatibility or metadata, this newly documented allowlist conflicts with skills/marius-mode/playbooks/authoring-a-skill.md:5, which says frontmatter must contain only the previous key set, while this file requires authors to follow that playbook. The validator accepts the new keys, so it will not expose the contradiction; update the playbook or remove these keys here so authors have one authoritative contract.
AGENTS.md reference: AGENTS.md:L21-L21
Useful? React with 👍 / 👎.
Brings AGENTS.md (which CLAUDE.md imports) up to date with what landed in #1-#7: Agent Skills spec limits, description rules for harnesses that ignore disable-model-invocation, the ALLOWED_TERMS escape hatch, shared references (bridge, transcripts), how to test skills and installs live without touching ~/.claude, the Copilot parse check, and the stacked-PR merge process. Stacked on #7.
🤖 Generated with Claude Code