The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external literature reviews and conversation history to update guide files, presenting a surface for indirect instructions. Ingestion points: Literature review folders (e.g., xxx/reviews) containing LaTeX (.tex) files. Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present. Capability inventory: Execution of a local validation script (scripts/validate_guide.py), file system write access to project guides, and optional network communication via the GitHub CLI. Sanitization: No specific filtering or sanitization steps are defined for the extracted insights before they are written to the guide file.- [DATA_EXFILTRATION]: The skill contains instructions to report bugs to the author's repository (huangwb8/bensz-bugs) via the gh tool. This is a legitimate developer feedback mechanism that includes strict safeguards, such as local staging of reports, mandatory desensitization of data, and the requirement for explicit user authorization before any network upload.- [COMMAND_EXECUTION]: The skill invokes a local Python script (scripts/validate_guide.py) and the GitHub CLI (gh). These operations are used for legitimate validation and reporting purposes and are constrained to the skill's defined functionality. The Python script performs safe string matching and does not use dynamic execution or shell spawning.

research-guide-updater