The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data by reading comments from GitHub issues in the pipecat-ai/provider-watch-reports repository to identify project decisions.
Ingestion points: Instructions in SKILL.md (Step 1.5) and RESEARCH_GUIDE.md (Step 0.3) describe how the agent should fetch and interpret these comments.
Boundary markers: No explicit markers are used to isolate the comments, but the agent is guided to extract specific decision-related keywords.
Capability inventory: The agent can perform file writes, create git branches, and execute localized scripts through uv run.
Sanitization: The skill enforces strict guardrails against credential leakage and requires manual review for publishing, mitigating the impact of potentially poisoned comments.
[EXTERNAL_DOWNLOADS]: The skill fetches data and code from external repositories.
Vendor Resources: The agent clones the official pipecat-ai/provider-watch-reports repository to manage project state.
Third-party data: The agent is directed to read performance data from the kwindla/aiewf-eval repository.

provider-research