Content
67%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 specialist workflow with an executable helper command, precise comparison semantics, explicit verdict labels, and appropriate externalization of the evidence contract and fixtures to real bundle files. Its weaknesses are dense, over-hedged prose with redundant prohibition lists repeated across sections, a Windows-specific path in the command, and a shared reference tree that chains beyond one level.
Suggestions
Consolidate the repeated prohibition/permission caveats ("separate appropriately approved workflow", "no new reads/jobs/traffic") into a single boundaries section stated once, cutting the redundant tokens that hurt conciseness.
Replace the Windows-style path "<skill-directory>\scripts\recovery.py" with a cross-platform form (e.g., <skill-directory>/scripts/recovery.py) so the command is copy-paste ready on any host.
Merge or shorten the "Standalone availability gate" and "Current-request intent gate" preamble into a compact entry checklist so the actual step 1-5 workflow sequence is immediately visible and traceable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | There is no concept-teaching padding, but prohibition lists repeat across sections: "no new tests, traffic or reads" in the entry section reappears as "not new reads, jobs, traffic or saving", and "separate appropriately approved workflow" appears in sections 2 and 5. Not 2 because all content is skill-specific rather than generic explanation; not 4 because the cross-section redundancy is more than minor and the prose could be tightened considerably. | 3 / 5 |
Actionability | Concrete command (`python "<skill-directory>\scripts\recovery.py" <sanitized-input.json>`), exact comparison semantics ("compare `total / sample_count` against the explicitly agreed bound in each window"), and a concrete verdict enum (VERIFIED_RECOVERED / NOT_RECOVERED / INCONCLUSIVE). Not 5 because input construction is deferred to contract.md and the path uses a Windows-only backslash; not 3 because the guidance is executable rather than pseudocode. | 4 / 5 |
Workflow Clarity | Numbered stages 1-5 clearly sequence scope agreement, evidence acceptance, comparison, reporting, and optional follow-up, with an explicit error feedback loop ("Invalid helper input instead produces a sanitized error/nonzero exit; correct it explicitly, never call it recovered"). Not 5 because the three preamble gate sections before step 1 obscure the entry sequence and some checkpoints are implicit; not 3 because validation checkpoints are largely present. | 4 / 5 |
Progressive Disclosure | The body appropriately externalizes detail to real, one-level-deep, purpose-signaled links ([the evidence contract](references/contract.md), [the fixture](references/fixture.json), [the fixture-p95 fixture](references/fixture-p95.json), scripts/recovery.py), all of which exist in the bundle. Not 5 because references/shared/next-actions.md chains onward to journey-indicator.md and handoff.md, and references/shared/investigation/ nests references and scripts two-plus levels deep; not 3 because references are clearly signaled, not buried, and the right content is split out. | 4 / 5 |
Total | 15 / 20 Passed |