feat: compose method skills within Trellis roles

[guanbear]

Summary

• add optional method_skills slots for brainstorm, implement, check, and debug
• resolve ordered project-local and global:<name> Agent Skills with canonical-path deduplication, boundary checks, visible diagnostics, and non-fatal fallback
• compose methods inside Trellis-owned roles across Markdown, TOML, JSON, native-hook, pull-based, inline, and channel-runtime entry paths
• add an additive 0.6.9 config migration, bilingual documentation, dogfood runtime parity, and integration coverage

Why

Trellis should remain responsible for task state, planning artifacts, lifecycle gates, Git constraints, dispatch, verification, and reporting, while reusable Agent Skills contribute role-specific methods such as grilling, tdd, codebase-design, code-review, and diagnosing-bugs. This avoids copying third-party prompts or replacing Trellis roles and losing workflow guarantees.

Behavior

method_skills:
  brainstorm:
    - global:grilling
  implement:
    - global:tdd
    - global:codebase-design
  check:
    - global:code-review
  debug:
    - global:diagnosing-bugs

Project-local references are directories relative to the repository root. Named global references resolve beneath ~/.agents/skills/. The first resolved method is primary, later methods are supporting, duplicate canonical paths are ignored, and invalid methods warn before Trellis continues with its built-in role.

This first version deliberately does not install, download, pin, discover, or recursively resolve skills, and it does not allow an external skill to replace a complete Trellis workflow role.

Validation

• Standards review: PASS
• Spec review: PASS
• generated-platform runtime audit: 20 platforms, 1,011 init/update files compared byte-for-byte, 32 implement/check role entries verified
• generated Python runtime probes: all four slots, project/global ordering, canonical deduplication, inline empty YAML containers, project-boundary symlink escape, global symlink-loop fallback, native-hook success and visible failure fallback
• Python py_compile: PASS
• template/dogfood Python files: byte-identical
• TypeScript syntax stripping: PASS
• migration manifest JSON parse: PASS
• git diff --check: PASS

The repository's Vitest/typecheck commands could not run locally because dependencies were absent and the restricted environment could not reach the npm registry. The added Vitest integration suite is intended to run in CI.

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 970bb181-cd0c-48dc-9efc-66f841be06fd

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}