The Agent Skills Directory

Local Command Execution: The skill executes a pre-flight check using python3 -m setup_preflight with a modified PYTHONPATH pointing to a project-local directory (.apache-magpie-local). This command is used to verify the skill's environment and configuration before processing the user's request, which is consistent with the skill's purpose as a developer tool.
Indirect Prompt Injection Surface: As a code review utility, this skill ingests untrusted external content (git diffs) that could potentially contain malicious instructions. The skill proactively addresses this by including explicit 'Injection Guard' instructions for its sub-agents, directing them to treat all diff content strictly as data and to report any directive-like text as a security finding.
Ingestion points: The skill collects code changes into a diff_text variable using git diff in Step 1.
Boundary markers: The instructions include explicit sections labeled 'Injection Guard' that define how to handle directive-like content.
Capability inventory: The skill uses local git commands and invokes the Agent tool to coordinate sub-reviews.
Sanitization: The skill relies on natural language instructions for the model to treat input as data and ignore embedded directives, which is a standard safety pattern for this type of task.

multi-agent-review