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.
The body is a highly actionable runbook with concrete, executable commands for every phase and a clear sequenced workflow with verification checkpoints and responsible-PoC guardrails. Main weaknesses are redundant, date-stamped bounty trivia that hurts token efficiency, npm-only code despite multi-registry claims, and a dangling upstream path reference.
Suggestions
Merge 'Programs that PAY' and 'Known exemplars' into one section and move the dated bounty figures into a clearly labeled reference, removing time-sensitive numbers from the main flow.
Add a verification step for the core precondition (confirm the target's CI actually resolves public registries for the internal scope) before publishing, and cover PyPI/Maven package builds or narrow the description to npm.
Fix the dangling 'skills/_corpus/payloads/Dependency Confusion/' cross-reference to point at a real file in the bundle or remove it.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with tables and executable code and assumes domain competence, but 'Programs that PAY' and 'Known exemplars' are redundant with each other and carry time-sensitive figures (2021 bounty amounts, '2022-2024') outside any deprecated/old-patterns section, which the guidelines penalize. Not 4 because these sections and the dated payout trivia need tightening; not 2 because most sections earn their tokens. | 3 / 5 |
Actionability | Copy-paste-ready guidance throughout: a curl pipeline for bundle recon, npm/PyPI/Maven name-availability checks, a complete package.json and beacon.js, and the publish command. Not 5 because the description promises npm/PyPI/Maven/Crates coverage but build-and-publish code exists only for npm, and placeholders ($TARGET, YOUR_INTERACT_DOMAIN) require substitution. | 4 / 5 |
Workflow Clarity | A clear numbered sequence (recon, verify, build, publish, observe, PoC framing, severity, defender) with checkpoints: step 2 is a go/no-go name-availability check, step 5 confirms execution via DNS callback, and the PoC section has explicit DO/DO-NOT guardrails plus cleanup. Not 5 because the core precondition — that the target's resolver searches both public and private registries — is asserted but never verified, a missing checkpoint for an outward-facing publish operation. | 4 / 5 |
Progressive Disclosure | Well-sectioned single-file skill with clear headers and a one-level-deep external reference (Birsan's writeup); no nested references. Not 5 because the ~150-line body exceeds the simple-skill exception and sections like defender mitigations and bounty exemplars could be split out or trimmed, and 'Cross-references' points to 'skills/_corpus/payloads/Dependency Confusion/' which does not exist in this bundle. | 4 / 5 |
Total | 15 / 20 Passed |