The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill uses the !command`` syntax to capture repository information (git status, diffs, and logs) into the context at load time. These commands are limited to standard, non-destructive git operations necessary for code analysis.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of git diffs and commit messages.
Ingestion points: Untrusted diff content and logs are ingested in SKILL.md and passed to sub-agents in WORKFLOW.md.
Boundary markers: Prompts utilize clear headers like DIFF CONTENT: and CLAUDE.md FILES: to delimit data, though it lacks explicit instructions to ignore embedded commands within the diffs.
Capability inventory: The skill has capabilities to write files to the local filesystem and post comments to GitHub PRs via the gh tool.
Sanitization: There is no explicit sanitization of the diff content, but the multi-agent 'Confidence Scoring' phase (Phase 3) acts as a high-signal filter to reduce noise and potential manipulation.
[COMMAND_EXECUTION]: The skill orchestrates the execution of local development tools including git, gh (GitHub CLI), and standard filesystem commands (mkdir). These operations are consistent with its stated purpose of reviewing code and managing reports.

code-review