The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses standard git commands (rev-parse, rev-list, diff, log) to compute scope and retrieve code changes. It also invokes the vendor's CLI tool (entire) to fetch session and checkpoint metadata. These commands are necessary for the skill's primary function of code review.
[REMOTE_CODE_EXECUTION]: No remote code execution or script downloads from external sources were detected. All execution is limited to pre-installed local tools (git, entire).
[DATA_EXFILTRATION]: The skill reads local repository data but does not contain any network operations to transmit this data to non-whitelisted domains. Findings are presented directly to the user within the agent session.
[PRIVILEGE_ESCALATION]: No commands requesting elevated privileges (such as sudo) or modifying system configurations were found.
[PROMPT_INJECTION]: The instructions do not contain markers attempting to override safety filters or extract system prompts. It uses structured headers like 'Entire Review:' for parsing purposes.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data (git diffs and transcripts) and presents a surface for injection. However, it follows a read-only audit workflow, does not auto-execute code, and uses specific review rules (references/review-rules.md) to maintain agent boundaries. The impact is assessed as low risk.

review