The Agent Skills Directory

[SAFE]: The skill defines a comprehensive and structured engineering review process. It provides clear guidance for the agent to follow a multi-step analysis including architecture, security, and test coverage.
[COMMAND_EXECUTION]: The skill uses the Bash tool for context gathering, specifically executing git and gh (GitHub CLI) commands in SKILL.md to identify branch details and PR information. These operations are restricted to metadata retrieval and are standard for development workflows.
[EXTERNAL_DOWNLOADS]: The skill utilizes the WebSearch tool to verify architectural patterns and research best practices. This is a legitimate research function intended to improve the quality of the engineering review.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze external technical plans and pull requests.
Ingestion points: Reads project files via Read and Glob tools, and command output from git diff and gh pr view via the Bash tool.
Boundary markers: Employs a rigid section-based review workflow and an 'Anti-shortcut clause' to prevent the agent from deviating from its instructions.
Capability inventory: Includes file modification (Write), shell access (Bash), and internet access (WebSearch).
Sanitization: Relies on a human-in-the-loop interaction model via AskUserQuestion, requiring user approval for each identified issue and recommendation before proceeding.

plan-eng-review