Content
85%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A well-structured, actionable skill body with a clear sequenced workflow and explicit validation/confirmation checkpoints. Its main weakness is redundancy: the closing Common Mistakes, Red Flags, and Always sections restate rules already embedded in the process steps.
Suggestions
Consolidate the repeated rules — "Do NOT stage docs/plans/*.md", "verify tests before options", and "no automatic worktree cleanup" each appear 3–4 times across the process, Common Mistakes, and Red Flags; state each once in the process and drop the duplications.
Trim the Common Mistakes, Red Flags, and Always sections to only net-new caveats not already covered in Step 1–4, cutting the body's token cost without losing guidance.
Replace the bare "<test command>" placeholder with a note to reuse the project's detected runner (already named at Step 1: npm test / cargo test / pytest / go test) so the re-verify step is copy-paste ready.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient imperative guidance, but several rules are restated 3–4 times — "Do NOT stage docs/plans/*.md" appears in Option 1, Option 2, Common Mistakes, and Red Flags, and "verify tests"/"no auto worktree cleanup" recur similarly. It is not level 3 because the Common Mistakes, Red Flags, and Always sections substantially duplicate the process steps rather than earning every token. | 2 / 3 |
Actionability | Provides real executable bash (git pull --autostash, git merge, git push -u origin, the worktree path extraction one-liner) plus exact menu and confirmation text. Placeholders like <feature-branch> denote genuinely variable values, so this is concrete and actionable rather than pseudocode — not level 2. | 3 / 3 |
Workflow Clarity | Clear Step 1–4 sequence with explicit validation checkpoints ("Stop. Don't proceed to Step 2" on failure, re-verify tests after merge) and a typed-confirmation feedback loop for the destructive discard. Not level 2 because checkpoints and error-recovery paths are explicit. | 3 / 3 |
Progressive Disclosure | No bundle files exist and the skill needs none; the body is self-contained with well-organized sections (Overview, Process, Quick Reference, Common Mistakes, Red Flags, Integration). Per the simple-skill note, well-organized sections without external references score 3. Not level 2 because structure and navigation are clear. | 3 / 3 |
Total | 11 / 12 Passed |