The Agent Skills Directory

[DATA_EXFILTRATION]: The skill reads sensitive Claude session history logs (.jsonl files) located in the user's home directory to generate session summaries. It also optionally transmits session digests to a central agent named 'jan' using the 'maw' tool as part of a fleet reporting protocol.\n- [COMMAND_EXECUTION]: The skill executes a local Python script (dig-session.py) included with the skill and utilizes various CLI tools such as git, gh (GitHub CLI), and maw. Actions that modify external state, such as creating GitHub issues, require explicit user approval through a confirmation prompt.\n- [INDIRECT_PROMPT_INJECTION]: The skill summarizes activity from previous session logs containing untrusted user input, creating a potential attack surface. 1. Ingestion points: Reads .jsonl session logs via the dig-session.py utility. 2. Boundary markers: The archivist sub-agent instructions do not include specific delimiters to ignore embedded instructions within the summarized text. 3. Capability inventory: The skill has permissions to write files, perform git operations, manage GitHub issues, and send messages via maw. 4. Sanitization: The dig-session.py script uses regular expressions and string slicing to extract relevant session content.

forward