The Agent Skills Directory

[COMMAND_EXECUTION]: Utilizes standard git and GitHub CLI (gh) commands to manage the development workflow.
Evidence: Commands include git checkout, git pull, git commit, git push, gh pr view, and gh api.
Scope: All operations are restricted to the brave/brave-core repository.
[PROMPT_INJECTION]: The skill demonstrates safe prompt engineering by enforcing strict behavioral constraints.
Evidence: Instructions such as "Only make changes the reviewer explicitly asks for" and "Do NOT make any additional changes" prevent the agent from exceeding the intended scope of work.
[PROMPT_INJECTION]: The skill ingests untrusted input from GitHub comments, creating an indirect prompt injection surface.
Ingestion points: Pull request details and review comments fetched in SKILL.md via gh commands.
Boundary markers: Not present; ingested content is not wrapped in protective delimiters.
Capability inventory: Executing shell commands (git, gh) and modifying local source files.
Sanitization: Not present; the agent parses raw comment text to identify actionable feedback.
[SAFE]: Employs robust user oversight for all high-risk operations.
Evidence: Mandatory confirmation is required before implementing changes, committing code, pushing to the remote repository, and posting comments back to GitHub.

impl-review