The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses dynamic context injection in the SKILL.md header to retrieve the current git branch and a summary of changes using git rev-parse and git diff. These are benign informational commands.
[COMMAND_EXECUTION]: The skill dynamically identifies and executes project-specific validation commands (e.g., pnpm check, cargo clippy) based on instructions found in the repository's CLAUDE.md file. This is standard and expected behavior for a developer productivity tool.
[PROMPT_INJECTION]: The skill ingests untrusted data from local source files and git diffs to generate its reports. While this presents a surface for indirect prompt injection, the risk is mitigated by the skill's requirement for explicit user approval before any automated fixes are applied.
Ingestion points: Reads all changed files and git diff output (Phase 2).
Boundary markers: No explicit delimiters or "ignore embedded instructions" warnings are used when passing content to the analysis agents.
Capability inventory: Modifies local files (Phase 5) and executes shell commands for project validation (Phase 1).
Sanitization: No sanitization or escaping is performed on the ingested code content.
[SAFE]: No signs of data exfiltration, obfuscation, or persistence mechanisms were found. The skill operates locally and maintains a human-in-the-loop workflow for all impactful actions.

polish