The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to 'Run all commands listed in prd.json config.quality_checks' during its autonomous loop. This creates a direct path for arbitrary shell command execution if the prd.json file is modified by an attacker or if the agent generates unsafe commands during its 'discovery' phase.
[REMOTE_CODE_EXECUTION]: The 'Pilot Mode' and 'Autonomous Loop' involve the agent independently generating, implementing, and testing code. This high-level autonomy, combined with the ability to execute shell commands for quality checks, presents a significant risk if the agent's task discovery process is targeted by indirect prompt injection from untrusted source code or documentation.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data to drive its behavior.
Ingestion points: Reads project state from prd.json, progress.md, and CLAUDE.md.
Boundary markers: None identified in the instructions; the agent treats the content of these files as state truth.
Capability inventory: The agent can perform git commits, update files, and execute arbitrary shell commands via the quality_checks configuration.
Sanitization: No evidence of sanitization or validation of the commands retrieved from prd.json before execution.
[EXTERNAL_DOWNLOADS]: The documentation references external resources such as the Ralph Wiggum GitHub repository and a methodology blog post. These are documented as informational origins for the skill's logic.

auto