The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted text from local editor session files (Sublime Text sessions and VS Code backups) which is then triaged by an LLM-based subagent.\n
Ingestion points: Editor state files are parsed in scripts/extract.py and saved to a local work directory.\n
Boundary markers: The subagent is guided by a specific triage prompt (references/triage-prompt.md) and required to return structured JSON, providing a clear schema for output.\n
Capability inventory: The skill can write files to the local filesystem (scripts/assemble.py) and move or modify application configuration files (scripts/reset.py).\n
Sanitization: The skill uses regex-based slugification for filenames to prevent path traversal and employs regex patterns to detect and flag secrets before they are processed by the LLM.\n- [COMMAND_EXECUTION]: The script scripts/rescuelib.py runs the system ps command to verify if editors are currently running. This is a local utility call used for operational safety to avoid modifying active session files.\n- [CREDENTIALS_UNSAFE]: The file tests/conftest.py includes a hardcoded dummy AWS API key (AKIAABCDEFGHIJKLMNOP) used strictly for testing the skill's ability to detect sensitive information.

editor-rescue