The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted findings and source files through sub-agents, creating a vulnerability surface for indirect prompt injection attacks.
Ingestion points: Findings are ingested from markdown report files, inline text, or pre-populated task lists. Sub-agents also read arbitrary source code files and context documents (SKILL.md, agent-prompt.md).
Capability inventory: The skill possesses significant capabilities including automated commits (tam-commit), code pushes (tam-push), and the ability to resolve GitHub PR threads via API calls (SKILL.md).
Boundary markers: There are no explicit instructions or delimiters defined in the sub-agent prompt (agent-prompt.md) to warn agents against following instructions potentially embedded within the findings or source code.
Sanitization: The skill does not perform sanitization of external content before it influences automated actions such as updating tasks or closing findings.
[COMMAND_EXECUTION]: The skill utilizes the GitHub CLI (gh) to perform operations on pull requests, including posting comments and executing GraphQL mutations to resolve review threads.
Evidence: Phase 3.4 of SKILL.md details the use of gh pr comment and gh api graphql to manage PR states based on the triage outcomes.

bmad-os-findings-triage