The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests session transcripts that may contain untrusted data from previous interactions or external tool outputs. * Ingestion points: scripts/session_digest.py reads JSONL files from the ~/.claude/projects directory. * Boundary markers: There are no explicit instructions or markers used to distinguish between trusted instructions and potentially adversarial content within the ingested transcripts. * Capability inventory: The skill can propose and facilitate changes to repository configuration files such as CLAUDE.md, .claude/settings.json, and Makefile. * Sanitization: The ingested content is parsed as JSON but not otherwise sanitized before being used to inform the agent's recommendations.
[COMMAND_EXECUTION]: The skill executes a local Python script scripts/session_digest.py to analyze transcript files. This script is part of the skill's package and is used to generate summaries of past session activity.

retro