The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from user interviews and local repository files (e.g., git history, documentation). It mitigates injection risks by enforcing a structured interview process with mandatory confirmation gates and explicit instructions to avoid fabrication or making assumptions about missing data.\n- [DATA_EXFILTRATION]: The skill analyzes repository metadata, including git churn and author history, to assess technical risk. These read operations are limited to the project environment and are used solely for generating discovery artifacts, with no evidence of unauthorized data transmission or credential harvesting.\n- [EXTERNAL_DOWNLOADS]: The skill mentions performing web searches for prior art research but includes provisions for offline environments. It emphasizes citing sources and does not perform unverified package installations or execute remote scripts.\n- [NO_CODE]: This skill consists of markdown-based instructions and templates without any embedded scripts or executables, reducing the overall attack surface.

tlc-discover