Content
71%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 well-structured runbook: the six-phase workflow is sequenced with explicit per-batch validation and fail-fast feedback loops, and bundle references are real, one level deep, and clearly signaled. The main weaknesses are duplicated gating/fail-fast sections and ambiguous placement of the pre-execution proof requirements relative to the phase sequence.
Suggestions
Deduplicate the fail-fast rule and the resolve/handles gate commands, which appear in both 'Failure mode'/'Invocation and Visibility Proof Requirements' and 'Baseline validation to confirm first' — merge them into one gating section.
State explicitly where the five invocation-proof commands run within the phase sequence (e.g., 'Phase 0: gates, before Phase 1') since both late sections claim to run 'first'.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body doesn't explain concepts Claude already knows, but it is noticeably redundant: the fail-fast rule appears verbatim in both 'Failure mode' and 'Baseline validation to confirm first', and the 'skills resolve'/'handles --check' gate commands are listed twice across 'Invocation and Visibility Proof Requirements' and the baseline section. This matches 'Mostly efficient but includes some unnecessary explanation or could be tightened'. | 3 / 5 |
Actionability | Concrete executable commands are given throughout (`./bin/ask repo status`, `rg`, `fd`, `git status`, five exact gating commands with flags, `./bin/ask repo validate --ephemeral`), plus a four-category decision matrix and an ordered batch plan, fitting 'Mostly executable guidance; concrete code or commands with minor gaps'. Not 5 because several directives remain high-level ('Run a short orientation pass', 'capture evidence, confidence, and risks') without a concrete pattern for what to record. | 4 / 5 |
Workflow Clarity | The six phases are clearly sequenced with explicit validation checkpoints and a real feedback loop ('stop at the first failed gate ... rerun the failed gate only after the blocker is fixed', 'Re-run relevant validation after each batch'), so the destructive/batch cap of 3 does not apply. Not 5 because the two post-workflow gate sections both claim to run before everything ('Baseline validation to confirm first', 'Before execution') yet appear after Phase 6 in document order, leaving ambiguity about where they sit in the sequence. | 4 / 5 |
Progressive Disclosure | The SKILL.md is a genuine overview with a dedicated 'Progressive Disclosure' section pointing one level deep to references/contract.yaml, references/evals.yaml, and references/task-profile.json — all of which exist in the bundle and match their stated purposes (machine-readable contract, benchmark/adversarial cases, evaluator thresholds). This matches 'Clear overview with well-signaled one-level-deep references; content appropriately split; easy navigation'. | 5 / 5 |
Total | 16 / 20 Passed |