The Agent Skills Directory

[COMMAND_EXECUTION]: The skill directs the agent to execute a variety of shell commands to manage the code review process. This includes using the GitHub CLI (gh pr) for retrieving metadata and submitting reviews, as well as npm and npx for running tests, linters, and coordinating the specialized agent swarm.
[INDIRECT_PROMPT_INJECTION]: The skill is primarily designed to ingest and analyze untrusted data from pull requests, which presents a surface for indirect prompt injection where malicious instructions embedded in code or PR descriptions could attempt to influence the agent.
Ingestion points: Source code files changed in the PR, pull request descriptions, commit messages, and metadata fetched via the GitHub API (SKILL.md, Phase 1 and 2).
Boundary markers: The workflow lacks explicit delimiters or instructions to treat PR content as untrusted data to prevent the agent from executing instructions found within the code being reviewed.
Capability inventory: The agent is granted capabilities to read the local filesystem (including a broad check of D:\Projects\*), spawn additional agents, and execute shell commands.
Sanitization: The workflow relies on external tools like ESLint, static analyzers, and the 'Security Manager' agent to identify issues, but it does not specify internal prompt-level sanitization for the data it processes.
[EXTERNAL_DOWNLOADS]: The skill utilizes npx to run the claude-flow utility. This execution pattern potentially downloads and executes packages from the public npm registry at runtime.

when-reviewing-pull-request-orchestrate-comprehensive-code-revie