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 well-structured, largely lean instruction skill with a clearly sequenced workflow, explicit validation checkpoints, and well-signaled one-level-deep references. Weaknesses are mild redundancy across Constraints/Validation/Failure mode/Gotchas, missing script invocation examples, and two broken asset links.
Suggestions
Deduplicate the repeated evidence-gap and evidence-invention rules (Validation/Failure mode, Constraints/Gotchas/Philosophy) into a single section, and link session-evidence-workflow.md only once, to lift conciseness from good to lean.
Add a one-line invocation example for each referenced script (e.g., the arguments scan_codex_sessions.py expects and a sample command), and show the shape of the keep/improve/merge/retire action table.
Fix the asset links: icon-small.png and icon-large.png point at ./agents/assets/, which is outside this bundle and doesn't exist; point them at the actual asset paths or drop them, and consider linking references/evals.yaml so it is discoverable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly lean, imperative bullets with no explanation of concepts Claude already knows, but there is repeated material: "stop and report the exact gap" appears in both Validation ("Fail fast: stop at first missing or unreadable evidence source and report the exact gap") and Failure mode; "Do not infer outcomes without evidence" (Gotchas) restates "Do not invent evidence" (Constraints) and "Evidence first" (Philosophy); and session-evidence-workflow.md is linked twice (lines 13 and 52). Fits anchor 4 (minor trims possible), not anchor 5 where every token earns its place, and well above the verbosity of anchor 3. | 4 / 5 |
Actionability | Concrete guidance throughout: named extraction scripts, exact collector root "~/.agents/session-collector", a bounded-snippet extraction rule, an explicit deliverable field ("schema_version: 1"), a root-cause taxonomy ("coverage gap, instruction drift, routing mismatch, or quality regression"), and ranking criteria. Minor gaps keep it below anchor 5: no invocation examples or arguments for the referenced scripts (they are thin runpy wrappers whose CLI is not shown), and the keep/improve/merge/retire action table format is described but never exemplified. Well above anchor 3, where guidance would be pseudocode or missing key details. | 4 / 5 |
Workflow Clarity | A clear six-step Procedure with a dedicated Validation checklist ("Verify each recommendation cites at least one concrete artifact", "Verify severity ordering is explicit and reproducible"), an explicit fail-fast checkpoint on missing evidence, and error-recovery guidance in Failure mode (stop and report the gap; request clarification on ambiguous scope). The batch-operation validation cap does not apply because validation steps are present. This matches anchor 5: sequenced steps, explicit validation, feedback for error handling. | 5 / 5 |
Progressive Disclosure | The "Read when:" signaling for references/contract.yaml and references/session-evidence-workflow.md is exemplary and one level deep (the workflow reference contains no nested file links), and the scripts are clearly listed with their purpose. Minor organization gaps keep it below anchor 5: the icon links "./agents/assets/icon-small.png" and "./agents/assets/icon-large.png" point outside the bundle and don't exist (assets/ contains only skill-refactor.png), session-evidence-workflow.md is referenced twice, and bundle materials like references/evals.yaml are never surfaced. | 4 / 5 |
Total | 17 / 20 Passed |