Content
75%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, highly actionable skill body with a copy-paste-ready command, exact output schemas, and strong validation semantics for degraded inputs. Its main weaknesses are repetitive 'never manually restore' warnings and duplicated confidence disclaimers across three contract paragraphs, and contract detail that belongs in the reference files rather than the main body.
Suggestions
Consolidate the three repeated 'never restore/override/fold rejected inputs manually' warnings and the duplicated confidence disclaimers into a single 'Degraded input rules' statement or checklist.
Move the detailed theme-input contract and provenance-gate specifics into references/exposure_framework.md (or a new reference), keeping only the flags and a one-line rule in SKILL.md.
Surface the auxiliary bundle scripts (evaluation_coverage.py, provenance_contract.py) in the Resources section so the full bundle structure is discoverable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Tables and the CLI command are efficient, but the same prohibition is restated three times across separate paragraphs — 'never restore its score or bias manually', 'Never override this degradation by manually copying...', 'Never fold a rejected theme into exposure manually' — and confidence disclaimers are duplicated ('Confidence measures input completeness, not freshness, verified quality or prediction accuracy'). Not 2: nothing explains concepts Claude already knows and most sections are tight; not 4: the theme/regime/verification contract paragraphs clearly need consolidation into one stated rule. | 3 / 5 |
Actionability | The Step 2 command is copy-paste ready with all eight input flags and output directory, the output section shows exact JSON and Markdown schemas, the recommendation table maps each enum to concrete portfolio actions, and the provenance section supplies exact flags with example values ('--run-id daily-20261017-unique --as-of 2026-10-17T13:00:00Z --max-age-days 3'). Matches the fully-executable anchor covering the common cases. | 5 / 5 |
Workflow Clarity | Steps 1-4 are clearly sequenced with substantial validation described: coverage 'verdict: complete' requirements, stderr warnings for rejected inputs, the provenance gate ('Failure stops before report output'), and explicit instruction to inspect 'inputs_provided', 'inputs_missing', 'inputs_unparseable' and 'input_warnings'. Not 5: validation is described as script behavior rather than agent-performed checkpoints, and there is no fix-and-retry feedback loop for gate failures; not 3: checkpoints are explicit and plentiful for this read-only analysis skill. | 4 / 5 |
Progressive Disclosure | The Resources section lists real, one-level-deep bundle files ('scripts/calculate_exposure.py', 'references/exposure_framework.md', 'references/regime_exposure_map.md' — all verified present) with one-line descriptions, and the body cross-references the framework doc. Not 5: the dense theme-input contract and provenance-gate prose are inlined in SKILL.md where they could largely live in references, and auxiliary bundle scripts (evaluation_coverage.py, provenance_contract.py) are not surfaced in Resources. | 4 / 5 |
Total | 16 / 20 Passed |