The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted WordPress block markup to perform structural validation. Malicious instructions could potentially be embedded in this markup to influence the agent's behavior.
Ingestion points: Markup is read from files or stdin in lib/validator.cjs.
Boundary markers: Documentation in SKILL.md instructs the agent to report all findings and avoid silent substitution of content.
Capability inventory: Limited to parsing via jsdom and official libraries; no network access is performed based on input content.
Sanitization: Relies on standard WordPress block parsing logic.
[REMOTE_CODE_EXECUTION]: The skill downloads and installs Node.js runtime dependencies upon first use.
Evidence: scripts/validate-block-markup.cjs runs npm ci to fetch packages.
Mitigation: The installation command uses --ignore-scripts to prevent the execution of arbitrary lifecycle scripts within the downloaded packages. All dependencies are pinned to specific versions from the official npm registry.
[DYNAMIC_EXECUTION]: The tool uses JSON.parse to evaluate attribute strings within the block markup.
Evidence: Found in lib/validator.cjs during block hydration to extract metadata from the serialized block comments.

validate-block-markup