Content
61%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-organized, safety-conscious catalog that respects token budget and gates real publication behind HITL approval, but it operates at summary altitude: no technique includes executable guidance, and only one of eight playbooks has a sequenced workflow. The single external link points outside the bundle and cannot be resolved.
Suggestions
Add concrete, executable guidance for at least the highest-value techniques — e.g., a sample gh CLI search query for pull_request_target workflows, or a concrete typosquat-candidate generation command — to lift actionability above summary level.
Sequence the main engagement workflow (select technique → dry-run → review report → HITL gate) rather than only the dry-run sub-flow, so all eight playbooks inherit an explicit checkpointed path.
Fix or remove the '../../red-team/tools-techniques.md' link, which lives outside this skill's bundle and is not resolvable by the skill loader.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly lean: table-driven playbook summaries, terse bullet lists, and no explanation of concepts Claude already knows (it never defines dependency confusion or SBOM). Minor padding keeps it from 5 — the 'grown 1,300% since 2020' growth statistic and the time-sensitive 'Most rewarding attack class in 2024-2026' line are not placed in a deprecated/old-patterns section per the time-sensitivity guideline. | 4 / 5 |
Actionability | Some concrete guidance exists — specific GH Actions misconfig patterns ('pull_request_target with actions/checkout of ${{ github.event.pull_request.head.sha }}'), the --dry-run flag semantics, and the supply_chain_real_publish=true gate — but nothing is executable: no commands, code, or scripts for any of the eight playbooks. It is above 2 because flags, YAML patterns, and scoring inputs (download counts + Levenshtein distance) are specific, but below 4 because the operating altitude is summary-only with no copy-paste-ready guidance. | 3 / 5 |
Workflow Clarity | The dry-run flow is a clear 4-step sequence ending in a human-review checkpoint, and the real-publish path has an explicit two-factor validation gate (ConOps flag AND HITL approval), so the destructive-path cap is satisfied. However, only one of eight techniques has any sequenced steps; the main engagement workflow (select technique → execute → report) is implicit, leaving validation checkpoints missing for most of the catalog. | 3 / 5 |
Progressive Disclosure | A well-organized single-file skill: clear section headers, a playbooks table, and an explicit disclaimer that techniques are inline rather than separately loadable, which avoids dangling internal references. It misses 5 because the intro links to '../../red-team/tools-techniques.md', a path outside this skill's bundle that the skill loader cannot resolve — the only reference in the file, and an unverifiable one. It is above 3 because structure and navigation are otherwise clean with no nested references. | 4 / 5 |
Total | 14 / 20 Passed |