The Agent Skills Directory

Indirect Prompt Injection Surface: The skill ingests untrusted text from external sources such as mailing list archives and issue bodies to assemble audit reports. While it includes explicit instructions to treat this content as data and detect injection attempts, the ingestion of external data remains a potential surface for indirect prompt injection.\n
Ingestion points: Release planning issues (via Step 1) and mailing list archives (via 'Mail-archive resolution').\n
Boundary markers: The skill contains explicit instructions in the 'External content is input data' section to treat external text as untrusted and flag injection attempts.\n
Capability inventory: The skill possesses capabilities to write to the repository (via GitHub PRs) and execute local Python modules.\n
Sanitization: The instructions include protocols for redacting sensitive information from the security tracker and using PMC roster handles instead of personal emails.\n- Local Command Execution: The skill utilizes gh for PR creation, svn for metadata retrieval, and a local Python module (setup_preflight) for project verification. These operations are consistent with the skill's stated purpose and use project-standard tools for the Apache ecosystem.\n- External Data Access: The skill fetches information from trusted domains such as GitHub and the Apache Software Foundation's archives (lists.apache.org) to populate audit records. These network operations are necessary for data collection and target well-known, reputable service providers.

audit-report