Content
85%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 instruction skill: a clearly sequenced protocol with real validation gates and rollback loops, concrete mechanisms and commands, and clean one-level-deep progressive disclosure into two verified reference files. The only weaknesses are mild — some duplicated links, run-on compound steps, and a few procedural steps lacking source specifics.
Suggestions
Consolidate the duplicated reference links (inline in 'Local preparation' vs. the 'References' section) into one location to save tokens.
Split the run-on Protocol steps 4, 7 and 8 into short sub-bullets so each rule is scannable rather than packed into one sentence.
Specify where correction signals are scanned from (e.g. transcript, learning log, review comments) to make the Extract step fully executable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and protocol-style with no explanations of concepts Claude already knows, but reference links are duplicated (inline in 'Local preparation' and again in 'References'), Protocol steps 4/7/8 cram multiple rules into single run-on sentences, and the 'Canonical response anchors' section adds marginal value — minor trims, matching anchor 4 rather than the fully lean anchor 5. | 4 / 5 |
Actionability | Guidance is concrete and executable for an instruction-only skill: an 8-label root-cause taxonomy, explicit improvement-item fields ('root cause, owner, chosen mechanism, verification/eval reference, adoption status'), a failure-shape-to-mechanism-to-verification table, and real commands (get_session_cost, prepare_retrospective, 'ags retrospective enable'). It stops short of anchor 5 because a few steps remain procedural, e.g. 'Scan for correction signals' does not specify which sources to scan. | 4 / 5 |
Workflow Clarity | The 8-step Protocol is clearly sequenced (Extract, Classify, Trigger Miss Check, Propose, Authorize, Evaluate, Review and promote, Log and report) with explicit validation checkpoints — held-out evaluation in step 6 and independent approval plus fresh eval evidence in step 7 — and feedback/rollback loops ('Pin approved versions, monitor regressions, revert to last verified version'), matching anchor 5. | 5 / 5 |
Progressive Disclosure | SKILL.md serves as a clear overview with a Structure tree, and both referenced files exist one level deep with well-labeled links ('Signal tables, taxonomy, report template'; 'Local preparation disposition, coverage, trigger, boundaries') including a working heading anchor (#trigger-miss-record resolves to the 'Trigger miss record' section in methodology.md) — matching anchor 5. | 5 / 5 |
Total | 18 / 20 Passed |