Review

Good concept — the resolve → learn → cleanup loop is a natural extension of the compound-engineering philosophy and the orchestrator pattern is the right architecture choice (compose over modify).

A few things to fix before merging:

1. Remove version bump and changelog entry (blocking)

The PR manually bumps plugin.json to 2.43.0 and adds a handwritten CHANGELOG entry. Per AGENTS.md, routine PRs must not do either of these — release automation handles it. Main is already at 2.44.0, so this would actually downgrade the version.

Remove:

• The plugin.json version and description changes
• The CHANGELOG.md entry

2. Fix naming: use hyphens, not underscores (blocking)

The new skill is resolve-todos-fully (hyphens) but it references and chains resolve_todo_parallel (underscores). These are sibling skills in the same namespace — they need to use the same convention. The plugin standard is hyphens (see every other skill: resolve-pr-parallel, ce-compound, git-worktree, etc.). resolve_todo_parallel is the outlier.

Rename resolve_todo_parallel → resolve-todo-parallel (directory and frontmatter name: field) as part of this PR, and update the reference in resolve-todos-fully accordingly.

3. Use semantic wording instead of slash reference for skill invocation

Per AGENTS.md, inside a pass-through SKILL.md, don't use slash syntax for skill references that won't be remapped cross-platform:

# Instead of:
Run `/compound-engineering:resolve_todo_parallel $ARGUMENTS`

# Use:
Run the `resolve-todo-parallel` skill with $ARGUMENTS

/ce:compound is fine since it's a published workflow command.

4. Make step 3 tool reference platform-agnostic

Step 3 says "Use the TaskList tool" — this is Claude Code-specific. Reword to describe the capability:

List all todos and identify those with `done` or `resolved` status, then delete them.

Everything else looks good: frontmatter, gates, summary output format, README/table updates, and the lfg/slfg integration.

Review (updated)

The resolve → learn → cleanup loop is a good idea. But this should be folded into the existing resolve_todo_parallel skill rather than being a separate orchestrator.

Why not a separate skill?

resolve-todos-fully is a thin wrapper — it calls resolve_todo_parallel, then ce:compound, then deletes todos. It has no independent use case (nobody would run step 2 or 3 without step 1), and it's only invoked from lfg/slfg which previously called resolve_todo_parallel directly.

A separate orchestrator makes sense when the composed skills are independently useful and the orchestration is a distinct workflow (like lfg composing ce:plan → ce:work → ce:review). Here the three steps are tightly coupled and always run together — that's just one skill with more steps.

Recommendation: Add the compound and cleanup steps directly to resolve_todo_parallel. That skill already owns "resolve todos" — extending it to learn and clean up is a natural evolution, not a new concern.

Other items still apply

1. Remove version bump and changelog entry. Release automation handles this. Main is already at 2.44.0 so the PR would downgrade the version.

2. Rename resolve_todo_parallel → resolve-todo-parallel while you're editing it. Hyphens are the plugin standard (see resolve-pr-parallel, ce-compound, git-worktree, etc.). Underscores are the outlier.

3. Use semantic wording for skill references inside the SKILL.md instead of slash syntax (per AGENTS.md cross-platform rules). /ce:compound is fine since it's a published workflow command.

4. Make tool references platform-agnostic. "Use the TaskList tool" → describe the capability instead.

lfg/slfg changes

If you fold everything into resolve-todo-parallel, the lfg and slfg references just need the rename update (resolve_todo_parallel → resolve-todo-parallel) — no workflow change needed since the skill itself now handles the full lifecycle.