Content
73%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.
An expertly structured orchestration skill with exceptional workflow sequencing, validation gates, and operational safety rules, all written for a capable agent without padding. Its main weaknesses are sheer inline bulk — the write-isolation promotion procedure and budget-ordering rules belong in companion files — and reliance on referenced companion files that are not present in this bundle.
Suggestions
Move the 11-step scratch-to-artifacts promotion procedure out of SKILL.md into a single companion file and reference it from both hunter and verifier prompts, removing the duplicated fenced block.
Condense the cost-budget section to its gate rule and spend order summary, delegating the step-by-step reservation arithmetic to a planning reference file.
Ship the referenced companions (HUNTING.md, RECONNAISSANCE.md, VALIDATION-AND-REPORTING.md, ATTACK-CLASSES.md, report-schema.json, validators) with the skill, or inline minimal stand-ins, since the body's phase instructions depend on them.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body avoids explaining known concepts and every rule is non-obvious operational policy, but at ~470 lines it inlines substantial detail (the 11-step scratch-to-artifacts promotion procedure, multi-paragraph budget spend ordering) that could be tightened or delegated to companion files, and it explicitly duplicates the promotion block across two referenced files. | 3 / 5 |
Actionability | Highly concrete instruction-only guidance — exact file names, JSON fields, regex for agent IDs, validator scripts, status strings, and two exact terminal states — with only minor gaps such as the abstract 'use equivalent platform capabilities' and hunting methodology delegated to companion files. | 4 / 5 |
Workflow Clarity | Six phases are clearly numbered in order with explicit validation checkpoints (schema validation, validate-findings.cjs, validate-coverage-ledger.cjs, fresh-agent verification), feedback loops (coverage-critic waves, budget gates with exact incomplete_reason values), and a hard 'never stop mid-phase' terminal-state rule. | 5 / 5 |
Progressive Disclosure | References are well-signaled and one level deep (RECONNAISSANCE.md, HUNTING.md, VALIDATION-AND-REPORTING.md, ATTACK-CLASSES.md, validators), but none of these companion files exist in the provided bundle, and a large amount of procedural detail remains inline in SKILL.md rather than split out. | 4 / 5 |
Total | 16 / 20 Passed |