jamiebrynes7/dotfiles/archive

Add a repo-local Claude Code skill at .claude/skills/task-implementer/SKILL.md that guides an agent through implementing a bean end-to-end.

The skill workflow:

1. Read the bean (beans show)
2. Validate the bean & flag discrepancies to the user for resolution before coding
3. Implement
4. Subagent code review (prefer critical-code-reviewer skill if available)
5. Mark the bean done + commit (code + bean file together)

Todos

• [x] Write SKILL.md with frontmatter (user-invocable) and the 5-step workflow
• [x] Verify discovery/format against existing repo skill conventions

Summary of Changes

Added .claude/skills/task-implementer/SKILL.md — a repo-local Claude Code skill that drives a bean from start to finish:

1. Read — beans show plus a GraphQL pull of parent/children/blockedBy for context.
2. Validate — a checklist (clear scope, unblocked, status fits, internally consistent, no stale assumptions) with a hard STOP that surfaces discrepancies to the user before any code is written.
3. Implement — follow the bean plan, check off todos as they land, run cargo test --workspace / nix flake check.
4. Subagent review — hand the diff to a fresh agent, preferring the critical-code-reviewer skill; triage by severity tier.
5. Done + commit — append a Summary of Changes, mark completed, commit code and bean file together using the repo's <area>: <summary> (id) convention.

Description follows the writing-claude-directives 'Use when...' discovery format; no allowed-tools restriction so the skill can use the full toolset.