writing-skills — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited writing-skills (Agent Skill) and scored it 45/100 (orange). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 2 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 2 flagged
A base64 string of 128+ characters appears in a documentation file. Encoded prompt injection hides the hostile instruction in base64 — invisible to keyword filters — and relies on the agent's ability to decode it at runtime. There is no normal authoring reason to embed a multi-hundred-byte base64 blob in skill docs.
*.sig, SIGNATURES) outside the documentation.A base64 string of 128+ characters appears in a documentation file. Encoded prompt injection hides the hostile instruction in base64 — invisible to keyword filters — and relies on the agent's ability to decode it at runtime. There is no normal authoring reason to embed a multi-hundred-byte base64 blob in skill docs.
*.sig, SIGNATURES) outside the documentation.Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Create the smallest reusable skill that changes decisions or improves task execution without constraining unrelated work. Skill changes use behavior validation, not a code TDD route. This validation does not require loading aegis:test-driven-development.
check already owns it. Prefer the smallest existing owner.
discovery tests, and bundled resources before changing or retiring them.
conventions in the project owner and automate mechanical constraints when practical.
Use representative tasks, not questions that merely ask the agent to recite instructions.
only when testing whether the skill itself needs to exist.
case appropriate to the skill type.
references, and unknowns. Do not treat line or byte counts as behavior proof.
If meaningful behavior cannot be evaluated, stop at needs-verification; do not claim that the skill is deployment-ready or capability-preserving.
criteria, safety boundaries, and steps whose omission caused an observed gap.
only for fragile, permission-sensitive, or safety-critical operations.
capability and when it applies; add an exclusion only when it prevents likely misrouting. Do not turn the description into a procedural shortcut.
SKILL.md.Move substantial conditional detail to a direct reference and state the evidence trigger for reading it.
by default or duplicate its content in the main body.
representative example is normally enough.
paths, and UI choreography in their host-owned documentation.
The canonical repository shape is skills/<skill-name>/SKILL.md with only the supporting references, scripts, or assets the workflow actually uses. Use the repository's host docs or the target host's live specification for optional metadata instead of copying a vendor manual into the skill.
Address observed gaps and remove superseded duplication in the same slice. Preserve public attribution and any externally supported contract. When retiring an internal owner, fallback, or resource, compose aegis:anti-entropy-governance; do not leave a compatibility copy without dependency evidence.
Do not convert one model's phrasing, one historical incident, or a hypothetical edge case into a universal rule.
Re-run the same positive, negative, and pressure/application scenarios under a comparable model, tool, and host configuration. Compare outcomes, not wording.
Also verify:
touched surface pass.
If behavior regresses, restore the missing obligation or narrow the change; do not add a second owner or fallback to mask it.
| Skill type | Representative evidence |
|---|---|
| Discipline or safety | realistic pressure and authorization boundaries |
| Technique | application, variation, and missing-information cases |
| Pattern | recognition, application, and counter-example cases |
| Reference | retrieval, correct use, and common-gap cases |
For a complex or high-risk skill where an independent evaluator adds confidence, read testing-skills-with-subagents.md.
Keep the result compact and auditable in natural prose or equivalent fields:
verified, needs-verification, or blocked.This is method-pack evidence, not authoritative completion or deployment approval.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.