The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's 'analysis' mode instructs the agent to 'verify the repro command if present' using grounding actions. This direction to execute shell commands extracted from untrusted bug analysis data (e.g., reproduction scripts in a bug report) creates a risk of arbitrary command execution if a malicious command is supplied by an attacker.
[PROMPT_INJECTION]: The skill processes untrusted external data from several sources, including plan.md, code diffs, and evidence files, to perform its analysis. It lacks explicit boundary markers or sanitization logic to prevent embedded malicious instructions in these files from overriding the agent's behavior (Indirect Prompt Injection).
Ingestion points: External files like plan.md, codebase source files (via Read/Grep), and analysis evidence strings.
Boundary markers: None identified in the instruction set for separating data from instructions.
Capability inventory: Includes file reading (Read), text searching (Grep), and shell command execution (Bash).
Sanitization: No validation or escaping of external content before processing is specified.

critical