The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill performs deep scans of codebase content to discover behavioral 'facts'. This ingestion of untrusted data from external code creates a surface for indirect prompt injection. Maliciously crafted code or comments could attempt to influence the agent's classification of facts or trick it into generating harmful validation commands.
Ingestion points: Codebase files, API contracts, and event shapes processed by the agent and its subagents.
Boundary markers: None provided; the instructions do not include specific delimiters or warnings to ignore embedded instructions within the scanned source code.
Capability inventory: The skill has access to shell command execution (via grep, jq, test), script execution (via node -e), and a specialized facts CLI tool.
Sanitization: No sanitization or validation of the content discovered in the codebase is specified before it is incorporated into the fact sheet.\n- [DYNAMIC_EXECUTION]: The process involves the agent dynamically generating shell commands and scripts (e.g., Node.js or Python snippets) to validate the truth of discovered facts. Because these commands are based on the agent's interpretation of the codebase, an attacker could potentially influence the generated command to perform unintended actions.\n- [COMMAND_EXECUTION]: The skill relies extensively on executing shell commands and a dedicated facts CLI to audit, list, and modify the project's fact sheet. This includes aliases for high-frequency operations like adding/removing tags and removing facts.

facts-discover