The Agent Skills Directory

[Indirect Prompt Injection Surface]: The skill is designed to ingest and process untrusted data from mailing list archives, including candidate names, email addresses, and desired Apache IDs.
[Evidence]: The skill documentation acknowledges this risk in 'Golden Rule 3', explicitly instructing the agent to treat external content as data rather than instructions. It provides guidance on detecting and flagging potential injection attempts (such as 'ignore previous instructions' strings) within these data fields.
[Mitigation]: The skill implements strong boundary controls, including mandatory redaction of third-party names via pii-redact, a human-in-the-loop requirement for all outbound communications ('Golden Rule 1'), and explicit validation logic for vote tallies.
[Tool Execution Context]: The skill invokes local tools and scripts for project configuration checks and privacy gating.
[Evidence]: Commands such as python3 -m setup_preflight and uv run ... privacy-llm-check are executed to ensure the environment is correctly configured and approved for processing potentially sensitive data.
[Context]: These operations are performed using local infrastructure within the project's framework, which is standard practice for maintaining environment consistency and data privacy.

committer-onboarding