Content
51%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 is maximally lean and its single reference is cleanly signaled, but it instructs almost nothing actionable: the analysis workflow is unspecified and the bundle's scan/correlate scripts and three other reference files are never pointed to. The skill currently reads as a stub overview rather than an executable guide.
Suggestions
Add a quick-start line that names the actual tooling, e.g. "Run `python scripts/scan_codex_sessions.py <sessions-dir>` to collect failures, usage, and coverage data", so the existing scripts are discoverable and the analysis has a concrete entry point.
Spell out the sequence with a validation checkpoint: scan sessions → identify failures/coverage gaps → rank recommendations, and require each recommendation to cite session evidence per the output_contract in references/contract.yaml before returning it.
Either add navigation pointers for the remaining bundle files (references/evals.yaml, references/task-profile.json) or state when they apply, so the whole bundle is reachable from SKILL.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is two sentences plus one conditional reference line ("Analyze skill reliability from session evidence and return prioritized recommendations.") with zero padding and no explanation of concepts Claude already knows. Every token earns its place, matching the lean-and-efficient anchor. Its brevity costs are charged to actionability, not here. | 5 / 5 |
Actionability | "Analyze skill reliability from session evidence and return prioritized recommendations" describes a goal rather than instructing how to achieve it — no command, code, or step for scanning or analysis. The only concrete guidance is the pointer to contract.yaml ("Read when: evidence schema or audit criteria are needed"), which is a minimal hint; the bundle's scripts (scripts/scan_codex_sessions.py, scripts/correlate_multi_source_skill_failures.py) are never mentioned, so it sits at the minimal-concrete-guidance anchor rather than entirely-vague (1) or some-incomplete-guidance (3). | 2 / 5 |
Workflow Clarity | The body implies at most a rough sequence (gather session evidence → analyze → return recommendations) but defines no steps, never sequences the two available scan/correlate scripts, and includes no validation checkpoint for what is a batch analysis over session history. It is above the incoherent/no-sequence anchor (1) only because a rough input-to-output flow is implied, and below anchor 3 because no checkpoints or explicit steps are written at all. | 2 / 5 |
Progressive Disclosure | The one reference present is well signaled and one level deep ("Read when: evidence schema or audit criteria are needed: [contract](./references/contract.yaml)", a real file), but scored against the actual bundle the navigation is incomplete: scripts/, references/evals.yaml, references/task-profile.json, and references/prompt-injection-expected-context.json are all unreferenced from SKILL.md, leaving the core tooling undiscoverable. That is more than the minor organization gaps of anchor 4, though the body itself is not a monolith, so it lands at 3. | 3 / 5 |
Total | 12 / 20 Passed |