The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The pull request review workflow in references/details.md ingests untrusted content from PR bodies, branch names, and author metadata. This data is used to determine if a contribution is AI-generated without robust sanitization, creating a surface for indirect prompt injection that could influence automated labeling or workflow logic.
Ingestion points: references/details.md reads github.event.pull_request.body and github.head_ref into the detect-agent-pr job.
Boundary markers: Absent in the scripts processing the external variables.
Capability inventory: The workflow utilizes the GitHub CLI with GH_TOKEN permissions to modify pull request labels (gh pr edit).
Sanitization: Absent; external content is evaluated directly via grep for detection patterns.
[EXTERNAL_DOWNLOADS]: Fetches and installs security utilities from well-known sources, including the git-secrets tool from the official Amazon Labs repository and the TruffleHog scanner from Truffle Security.
[COMMAND_EXECUTION]: Provides administrative helper scripts that execute system-level commands, such as useradd, setfacl, and iptables, to configure restricted workspace environments and network controls.

ai-coding-agent-guardrails