The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted code content from pull requests and commits.
Ingestion points: Phase 0 uses gh pr diff and git diff-tree to ingest external code changes.
Boundary markers: Instructions in references/agent-prompts.md direct the agent to confine scans to the provided file list and hunk ranges.
Capability inventory: Utilizes gh, git, grep, and subagent tools for analysis.
Sanitization: No explicit sanitization is described for the ingested diff data.
[COMMAND_EXECUTION]: The skill executes local shell commands to retrieve code diffs for analysis.
Evidence: SKILL.md contains specific Phase 0 commands gh pr diff and git diff-tree -p to facilitate the audit.

review-perf