Content
81%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.
A highly actionable, tightly sequenced workflow skill: exact commands, verbatim templates, explicit gates, and strong per-step validation make it nearly impossible to execute wrongly. Its one real weakness is token efficiency — repeated rule statements and measured-failure anecdotes inflate a body that its own reference file says was slimmed to fit the read-time budget.
Suggestions
Trim the measured-failure anecdotes to one-line calibration facts or move them to a reference file (e.g. the comment-delta "three traps" cases, the +66/+85/+285 examples, and the "attack map with a timer" prose in Step 6).
State each repeated rule once and cross-reference it: the six-section shape rule appears in the preamble, Step 8, and the shape-check section, and the TASK-BRANCH/collector attribution rationale is given in full in both Step 4 and Step 6.
Reference registering-a-stack.md directly from Step 9 (where stack registration is discussed) instead of routing through workflow-steps.md, and consider moving the long draft-or-ready decision rules of Step 9 into a reference file to match the pattern used for Steps 3/5/7.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~6,500-word body never teaches known concepts and is mostly dense repo-specific rules, but it repeats key rules verbatim in multiple steps (the six-section shape rule in the preamble and Step 8; the TASK-BRANCH collector rationale in Steps 4 and 6) and carries several anecdotal case studies ("added 66 comment lines net", "+85 for a branch actually running -40", "+285/-9 reported for +187/-0") that could be trimmed or moved to references. It sits between "noticeably verbose" (2) and "minor over-explanation" (4) — the padding is real but confined to justification prose, so it lands at the midpoint-adjacent 3. | 3 / 5 |
Actionability | Fully copy-paste-ready throughout: exact bash for base resolution, changeset checks, and the grep -c '^## ' shape check; verbatim PR-body and report skeletons; exact labels, repos, and --type values; and agent dispatch prompts with the required TASK-BRANCH: opener — with explicit decision tables (full/trivial-tests/trivial) covering the common cases. | 5 / 5 |
Workflow Clarity | Steps 0–10 are strictly ordered with explicit validation checkpoints and feedback loops at every risky point: "Both must print 6", "The push waits until it prints nothing", re-run gates after fix rounds, the **Review complete.** marker for partial reviews, and stop-before-return rules for dispatched reviewers. | 5 / 5 |
Progressive Disclosure | Five real reference files are clearly signaled and opened per-step ("Open this file when you reach the step"), but the structure is not uniformly one level deep: registering-a-stack.md is reachable only through workflow-steps.md (two levels), never referenced directly from the body. Good placement overall with a minor organization gap — not 5. | 4 / 5 |
Total | 17 / 20 Passed |