The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a workflow to parse external Markdown documents to extract metadata (IDs and titles), which creates a vulnerability surface where a downstream agent might execute instructions embedded within those document headings. * Ingestion points: Markdown files located in specs/ and specs/external-requests/ directories. * Boundary markers: The manifest generation workflow does not define specific delimiters or warnings to isolate extracted content in the final manifest. * Capability inventory: The skill performs file reading, regex parsing, and SHA-256 calculation; no direct command execution or network access is defined. * Sanitization: No evidence of input validation or sanitization for the content of extracted headings before manifest generation.
[METADATA_POISONING]: The skill package includes a skill-report.json file containing a detailed self-assessment and security audit that claims the skill is 'safe'. These claims are self-reported metadata and are evaluated as data points rather than authoritative security conclusions.

spec-discovery