The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from an external source which could potentially contain malicious instructions intended for the AI agent.
Ingestion points: The h1_reference.py script fetches content from disclosed reports via the HackerOne GraphQL endpoint.
Boundary markers: Absent; the skill does not appear to use specific delimiters or instructions to prevent the agent from obeying instructions embedded within the fetched report content.
Capability inventory: The skill allows for the execution of shell scripts (smoke.sh, sync-skill-content.sh) and Python scripts (secret_scan.py, h1_reference.py).
Sanitization: Absent; there is no evidence that the external content fetched from HackerOne is filtered or sanitized before being presented to the agent.
[DYNAMIC_EXECUTION]: The driver script employs dynamic script generation to perform metadata validation.
Evidence: The smoke.sh script utilizes a heredoc (`python3
"$skill" <<'PY' ... PY) to generate and execute an inline Python script for validating YAML frontmatter in SKILL.md` files.
[COMMAND_EXECUTION]: The skill is designed to facilitate the execution of local shell and Python scripts as part of its build and test process.
Evidence: The skill documentation and the smoke.sh driver explicitly invoke secret_scan.py, h1_reference.py, and sync-skill-content.sh to validate the environment and repository state.
[SAFE]: The skill includes hardcoded security tokens for functional testing, but these are identified as well-known CI canaries.
Evidence: The smoke.sh script and SKILL.md documentation include a dummy AWS Access Key (AKIAIOSFODNN7EXAMPLE) and a standard test JWT token to verify that the secret_scan.py helper is working correctly.

run-claude-osint