The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local git commands to identify changes, including git merge-base, git diff, and git ls-files. These are standard operations for the skill's stated purpose of reviewing development branches.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the codebase being reviewed, which could contain malicious instructions.
Ingestion points: The agent reads the cumulative diff from /tmp/review_branch.diff and the contents of untracked files identified in /tmp/review_branch_new.txt.
Boundary markers: The instructions do not mandate the use of XML tags or clear delimiters to separate the diff data from the agent's instructions, increasing the risk that instructions inside code comments could be followed.
Capability inventory: The skill has the capability to write to the file system (to fix issues found) and execute shell commands via the git toolset.
Sanitization: There is no mention of sanitizing or filtering the content of the diffs before they are processed by the review agents.

review-branch