feat: rationalize todo skill names and optimize skills#368
Conversation
…dance The old rule was overly broad. The real issues are: box-drawing chars breaking markdown tables across platforms, and non-ASCII in identifiers breaking converter regex patterns. Emoji and Unicode punctuation in prose/skill content are universally supported and fine to use.
Rename todo-related skills to a unified `todo-*` namespace: - file-todos -> todo-create - triage -> todo-triage - resolve-todo-parallel -> todo-resolve Trim skill content for conciseness and resilience: todo-create reduced from 231 to ~100 lines, todo-triage from 312 to ~90 lines. Skills now describe intent and workflow rather than prescribing rigid templates, letting agents adapt their output naturally. Update all cross-references in 10 consuming skills/files, converter, codex-agents, tests, and README.
todo-resolve now only resolves `ready` todos, skipping `pending` ones and reporting them at the end. This makes the pending/ready distinction load-bearing — pending todos require triage before automated resolution. Adds a solution doc capturing the design decision and the relationship between ce:review, ce:review-beta, and the todo status lifecycle.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: ae0bb2ddd3
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".
| 5. `/ce:review` | ||
|
|
||
| 6. `/compound-engineering:resolve-todo-parallel` | ||
| 6. `/compound-engineering:todo-resolve` |
There was a problem hiding this comment.
Insert todo-triage before todo-resolve in LFG workflow
Step 5 runs /ce:review, which writes findings as pending todos, and step 6 now immediately runs /compound-engineering:todo-resolve; however todo-resolve explicitly skips pending items and only processes ready ones. That means the post-review fix phase silently does no work for newly created findings, so this “full autonomous” sequence no longer resolves review output unless a triage gate happens first. Add a triage step (or switch to ce:review-beta mode:autonomous, which emits ready todos) before todo-resolve; the same ordering problem appears in skills/slfg/SKILL.md.
Useful? React with 👍 / 👎.
Summary
todo-*namespace:file-todos->todo-create,triage->todo-triage,resolve-todo-parallel->todo-resolvePerformance improvements
Ran old-vs-new skill comparisons via subagents across 3 eval scenarios (create todo, trivial fix triage, standard triage):
The triage improvement comes from the old skill's heavily templated approach causing agents to reproduce verbose boilerplate (re-presenting entire items on custom edits, explaining what pending->ready means, etc.). The new skill describes the workflow's intent and lets agents adapt their output naturally.
Test plan
bun testpasses for review-skill-contract, release-metadata, codex-agentsbun run release:validateconfirms 43 agents, 41 skills, 1 MCP server