The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's 'Self-healing validation' phase instructs the agent to automatically execute build/compile commands and run project test suites. Executing arbitrary project code autonomously is a high-privilege action that should be reviewed before use.
[PROMPT_INJECTION]: The skill explicitly instructs the agent to be 'autonomous' and 'Do NOT ask the user questions.' This instruction, combined with active command execution capabilities, bypasses human-in-the-loop safety controls and could lead to unintended actions if the agent encounters malicious instructions in the data being audited.
[PROMPT_INJECTION]: The 'Self-evolution telemetry' section directs the agent to access and append data to files in ~/.claude/projects/. While intended for tracking execution metadata, this involves automated file system operations on sensitive configuration paths.
[PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection (Category 8):
Ingestion points: The skill processes user-supplied $ARGUMENTS and external 'housing management system' data (unit inventory, tenant data, regulatory reports) across four phases.
Boundary markers: None identified. There are no delimiters or instructions to ignore embedded commands within the ingested data.
Capability inventory: The agent has the capability to execute shell commands (build/test), write to the local file system (~/.claude/projects/), and invoke other skills.
Sanitization: No sanitization or validation of the input data is described before it is processed by the agent's core logic.

housing-audit