Repository navigation
docs: tstack guide, README, and final review fixes - #7
Conversation
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Settle the push/PR/fan-out grant in one rule, make delegation optional in Feature, use installed skill paths instead of tstack checkout paths, replace /goal and cloud-sleeper with harness-neutral mechanisms, narrow slash-only skill descriptions, route cross-family review through the bridge, and write verification skills to .agents/skills. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
check.mjs now enforces the spec's name charset and length, the 1024 character description limit, accepts compatibility and metadata, and strips YAML quotes. Copilot CLI's own parser lists all 45 skills. 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: eafcf4cfcb
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".
|
|
||
| export const PORTABLE_KEYS = new Set(["name", "description", "license", "allowed-tools", "disable-model-invocation"]); | ||
| export const PORTABLE_KEYS = new Set([ | ||
| "name", "description", "license", "compatibility", "metadata", "allowed-tools", "disable-model-invocation", |
There was a problem hiding this comment.
Including compatibility and metadata in PORTABLE_KEYS makes node scripts/check.mjs accept skills that violate this repository's frontmatter contract; the authoring playbook also still instructs authors to use only the existing five optional/required fields. Keep these keys rejected, or update the repository contract and every authoring reference as part of the same change.
AGENTS.md reference: AGENTS.md:L14-L14
Useful? React with 👍 / 👎.
| --- | ||
| name: principle-build-the-lever | ||
| description: "Apply to any non-trivial work, not just bulk work: edits, migrations, analyses, checks. Build the tool that does it or proves it (codemod, script, generator, or a skill your subagents follow) instead of working by hand. The tool is the artifact a reviewer can rerun." | ||
| description: "Apply to non-trivial, repeated work: edits, migrations, analyses, checks. Build the tool that does it or proves it (codemod, script, generator, or a skill your subagents follow) instead of working by hand. The tool is the artifact a reviewer can rerun. One-off operational work takes the simplest direct path instead." |
There was a problem hiding this comment.
For one-off non-trivial work, the new description says to take the simplest direct path, while this same skill's body still says to build a lever for every non-trivial task, skip only trivial tasks, and explicitly states that repetition is not the bar. An agent that loads the skill directly therefore receives the opposite instruction from the applicability contract; update the body to reflect the new exception or retain the prior description.
AGENTS.md reference: AGENTS.md:L17-L17
Useful? React with 👍 / 👎.
Rewrites the guide and README for tstack (per-harness install via npx skills, bridge setup, marius-mode gates). Fixes all 8 Important findings from the whole-branch review and adds Agent Skills spec limits to check.mjs for Copilot compatibility (copilot skill list parses all 45 skills).
Verified: check.mjs ok (45 skills), validator tests 14/14, marius-mode scripts 52/52, strict typecheck, npx skills install into clean homes for Claude Code, Codex, Copilot, Cursor with exec bits intact, live /marius-mode run routed to Investigation.
Work-machine checklist (Copilot CLI):
Stacked on the marius-mode PR.
🤖 Generated with Claude Code