infer-conventions
Pass
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill analyzes local codebase files to generate persistent instructions for AI agents.\n
- Ingestion points: The skill reads various project files including composer.json, linting configurations, and PHP files within the app/, routes/, and tests/ directories.\n
- Boundary markers: No explicit boundary markers or delimiters are defined for data ingestion; however, a mandatory human-in-the-loop confirmation step is required before any inferred conventions are saved.\n
- Capability inventory: The agent uses file system inspection tools (ls, grep) and the record-rule tool to write to the .ai/rules directory.\n
- Sanitization: Instructions explicitly require the agent to strip all detection evidence, file paths, and code snippets from the final output, minimizing the risk of propagating malicious content from the codebase into agent instructions.\n- [COMMAND_EXECUTION]: The skill relies on standard shell utilities to perform static analysis.\n
- Evidence: The instructions guide the agent to use ls to map the directory structure and grep to search for specific Laravel patterns and syntax within the codebase. These actions are appropriate for the skill's analytical purpose and are limited to local filesystem inspection.
Audit Metadata