The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by processing project documentation and source code through AI agents.
Ingestion points: Markdown files and codebase source files are read by the script and provided to subagents (Dependency, Behavioral, and Code Example verifiers) in Phase 2b.
Boundary markers: There are no explicit delimiters or instructions provided to the agents to prevent them from following commands embedded within the documentation or code files.
Capability inventory: The skill uses Bash, Grep, Glob, and Read tools.
Sanitization: Input for internal shell commands (like grep) is escaped using re.escape().
[EXTERNAL_DOWNLOADS]: The verify_claims.py script performs network reachability checks on URLs found in documentation.
Evidence: The verify_url function uses urllib.request.urlopen with the HEAD method to check if external links in the documentation are active.
[COMMAND_EXECUTION]: The skill uses subprocesses to execute system commands for project analysis.
Evidence: scripts/verify_claims.py and scripts/extract_claims.py call git and grep via subprocess.run to determine project roots and search for code symbols.

doc-claim-validator