Content
61%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.
The body delivers unusually concrete operational guidance — exact timeouts, commands, sentinel formats, and a ready-to-use template — and handles destructive checkout operations with explicit validation. Its weaknesses are structural: the lifecycle is scattered across parallel bullets rather than sequenced, and the checkout-strategy and verdict-format instructions are restated multiple times, spending tokens on repetition.
Suggestions
Reorder the bullets (or add a short numbered lifecycle: turn start → before ending → PR handoff → review → verdict → checkout cleanup) so the sequence and its checkpoints are explicit rather than inferred.
State the checkout strategy once and reference it from the gate ask template instead of repeating it in the bullets, the template, and the closing line.
Add a brief in-line gloss or hyperlink context for internal jargon like "the row" and "the press", or replace those terms with self-describing ones.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly dense, high-value specifics (exact env var `COMMONLY_AGENT_RUN_TIMEOUT_MS`, value `2700000`, the SIGTERM consequence, exact `GATE: PASS`/`GATE: HOLD` sentinels), but the checkout-strategy guidance is repeated roughly four times (two bullets, inside the gate ask template, and a closing line) and the rules-44/51 verdict format is specified twice. Anchor 3 ("mostly efficient but... could be tightened") fits; anchor 4 is blocked by that repeated material. | 3 / 5 |
Actionability | Guidance is largely executable: specific timeout values and timing targets, `gh pr ready` behavior, the exact sentinel format with first-line/final-line placement rules, and a copy-paste-ready gate ask template. It falls short of anchor 5 only due to minor gaps — unexplained jargon ("the row", "the press") and no concrete commands for "commit and push what you have". | 4 / 5 |
Workflow Clarity | An implicit lifecycle exists (end turn early → open ready PR → one self-review round → answer CodeRabbit/get CI green → file verdict → checkout cleanup), and validation for the destructive checkout operations is explicitly present ("inspect... after checking for uncommitted work"; "Never use `git worktree remove --force` before checking"), so the destructive-operation cap does not apply. However, the steps are parallel bullets rather than a sequenced workflow, and checkpoints like CI-green are stated as rules rather than ordered gates — anchor 3 ("Steps listed but... checkpoints missing or implicit"). | 3 / 5 |
Progressive Disclosure | The skill is under 50 lines with no bundle files, the gate ask template is split into its own clearly-purposed section, and the external review-checklist references (rules 44 and 51) are well signaled and one level deep. It misses anchor 5 because the main bullet list interleaves turn management, PR etiquette, review rounds, and checkout hygiene without grouping or sub-headers — minor organization gaps per anchor 4. | 4 / 5 |
Total | 14 / 20 Passed |