The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses git diff --name-only to identify changed files and git rev-parse --abbrev-ref HEAD to derive review identifiers from the current branch.- [DYNAMIC_EXECUTION]: The skill executes a local Python script gen_run_stamp.py using uv run to generate unique, collision-resistant identifiers. It also dynamically spawns sub-agents via the Agent() interface with task-specific instructions to claim and execute tasks within a SAM plan.- [INDIRECT_PROMPT_INJECTION]: The skill processes code diffs which constitute untrusted data. It implements strong architectural mitigations:
Ingestion points: git diff output is used for file listing and as the target for sub-agent reviews.
Boundary markers: Review tasks are isolated within a structured SAM plan, with results written to specific task sections.
Capability inventory: Access is restricted to git commands and MCP-based plan operations.
Sanitization: The skill performs sanitization of branch names and enforces strict reconciliation checks (Check 6 and Check 7) to verify that the final synthesized punch list accurately reflects the raw verdicts from the security, performance, quality, and accessibility reviewers.

multi-perspective-review