Content
77%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
An unusually action-dense skill: concrete commands, exact paths, and explicit error-recovery rules for every refusal path make it fully operational. Its weaknesses are structural — a monolithic single-file layout with no reference files, and verbose rationale prose (including a version-pinned aside) that inflates the token cost of every invocation.
Suggestions
Move hook and ledger implementation internals (lock mechanics with CARNET_LOCK_WAIT, cache paths, transcript-entry/payload details, the 'as of Claude Code 2.1.276' version note) into a one-level-deep reference file such as references/hooks.md, keeping SKILL.md to the rules, commands, and how to read each hook message.
Tighten rationale prose: replace nested-clause explanations (e.g. the 'Without that, a session that wrote "comment on carnet#N" into its own wakeup…' sentence) with one-line cause→effect statements, and drop the duplicated peer-message handling explained both in 'What the hooks do for you' and in 'A peer naming an issue is not assigning it'.
Relocate time-sensitive version references into a dedicated compatibility note or reference file so the main body does not carry version numbers that will go stale.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Every line carries project-specific operational fact (no explanations of concepts Claude already knows), so it is above 2 — but the maximalist prose could be tightened throughout: long nested-clause rationales like "Without that, a session that wrote 'comment on carnet#N' into its own wakeup would hold #N for as long as it ran, the user never having typed the number", hook behavior explained twice (once in "What the hooks do for you", again in "A peer naming an issue is not assigning it"), and a version-sensitive aside ("as of Claude Code 2.1.276 the payload carries no source yet"). Mostly efficient but padded in places — the 3 anchor. | 3 / 5 |
Actionability | Fully executable, copy-paste-ready guidance: the Commands block gives exact invocations ("$C claim 12", "$C close 12 --why \"…\" --commit <sha>", "$C label 12 +critical -bug"), a verbatim comment-reading command ("gh api \"repos/$($C tracker)/issues/12/comments\" --paginate -q '.[].body'"), the script path, exit codes, --dry-run, and --steal semantics. Common cases are all covered. | 5 / 5 |
Workflow Clarity | The rules give a clear claim → work → release/close sequence with explicit validation checkpoints and feedback loops for every failure mode: exit code 2 → "Tell the user who and which session, and stop"; "carnet: could NOT claim" → run claim yourself with three retry attempts; unreadable comments → "exit 1 and change nothing"; SessionEnd as backstop. Destructive-ish operations (--steal) are explicitly gated to the user's decision, so no validation cap applies. | 5 / 5 |
Progressive Disclosure | The bundle contains no references/, scripts/, or assets/ directories — everything, including implementation internals (ledger lock mechanics with CARNET_LOCK_WAIT, cache paths, hook payload/transcript details, sender message templates), is inlined in one ~220-line SKILL.md that loads fully on every trigger. Section headers are clear (above 2), but content that would sit better in a one-level-deep reference file is inline, which is the 3 anchor rather than 4. | 3 / 5 |
Total | 16 / 20 Passed |