authoring-verifiable-skills
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [SAFE]: The skill serves as a meta-instructional guide for authoring other agent skills. It does not contain any malicious instructions, credential exfiltration, or unauthorized network operations. Its primary function is to enforce quality and security standards for skill development.
- [COMMAND_EXECUTION]: The skill instructions reference local verification tools (e.g.,
node tools/verify-skill.mjs) used for structural invariant checks and mutation testing. These operations are limited to the local environment and the skill's own directory structure, targeting structural validation rather than arbitrary command execution. - [INDIRECT_PROMPT_INJECTION]: The skill defines a framework for processing external skill files while mitigating injection risks.
- Ingestion points: The agent processes user-provided skill files (SKILL.md, rules/*.md) to perform authoring or repair tasks.
- Boundary markers: The skill mandates a 'Routing Gate' table in SKILL.md to clearly associate observable evidence with specific rules, preventing unintentional execution of instructions within data.
- Capability inventory: The skill performs file reads and executes local structural validation scripts via Node.js to ensure skill integrity.
- Sanitization: The skill enforces the 'Portable By Default' rule to exclude sensitive environment data (paths, URLs, credentials) and requires a strict parser for frontmatter to prevent interpretation of malicious metadata.
Audit Metadata