The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes Git CLI commands to gather repository state and generate code patches.
Evidence: The scripts/prepare_review_bundle.py script contains functions run_git and git_output which invoke subprocess.run to call git rev-parse, git status, and git diff.
Context: These calls are implemented using a list of arguments rather than a single shell string, which is a secure pattern that prevents command injection via repository names or branch references.
[PROMPT_INJECTION]: The skill facilitates an 'Isolated Review' workflow that is susceptible to indirect prompt injection.
Ingestion points: The scripts/prepare_review_bundle.py script aggregates content from the Git working tree, user-provided request files, and writer summaries into a bundle directory.
Boundary markers: The reviewer-prompt.md template and references/isolation-protocol.md lack robust boundary delimiters or explicit instructions for the reviewer agent to ignore potential instructions embedded within the code diffs or summaries.
Capability inventory: The skill is configured with powerful tools including Write, Edit, and Bash. If an agent is successfully manipulated by a malicious payload in a code diff, it could use these tools to perform unauthorized actions.
Sanitization: There is no evidence of sanitization or filtering of the content being bundled for review.

code-reviewer