The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a central Bash script to coordinate complex review tasks, including running test suites and mutation testing tools. These operations are performed within the task's worktree using a managed Bash environment.
[DYNAMIC_EXECUTION]: Shell commands are dynamically retrieved from external 'recipe' files associated with specific frameworks (e.g., Python, Node.js). These commands are resolved and executed at runtime, which is a functional requirement for a generic review tool but constitutes dynamic execution.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection because it ingests and processes untrusted data from the project being reviewed:
Ingestion points: The skill reads finished.json, snapshot.json, code diffs, research findings, and external recipe files into the agent's context.
Boundary markers: Instructions explicitly forbid the main agent from reading or pasting large file bodies into the conversation, directing the sub-agent to access them on disk instead. This limits the primary agent's exposure to potentially malicious payloads in the code diffs.
Capability inventory: The skill possesses the capability to execute shell commands via Bash and to dispatch an 'architecture-reviewer' sub-agent that has broad read access to the filesystem.
Sanitization: The skill implements validation for placeholder values using pc_refuse_forged_value and relies on internal platform libraries to handle command resolution.

review