The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process historical conversation logs, which serve as an untrusted data source that could contain adversarial instructions.\n
Ingestion points: Reads conversation history from JSONL files located in ~/.claude/projects/ via extract-conversations.py.\n
Boundary markers: Messages are prefixed with [USER]: and [ASSISTANT]:, but the skill lacks robust delimiters or explicit instructions for the analysis agents to ignore embedded commands within the text.\n
Capability inventory: The skill can write to the brain/ directory and possesses the ability to directly modify the SKILL.md instruction files of other skills based on its findings.\n
Sanitization: No sanitization or filtering is performed on the content of the extracted messages beyond simple truncation.\n- [DATA_EXFILTRATION]: The skill accesses highly sensitive user data stored in the local file system.\n
Sensitive file paths: It reads the full archive of past Claude conversations from ~/.claude/projects/, which may contain private information, credentials, or proprietary code discussed in previous sessions.\n- [COMMAND_EXECUTION]: The skill executes local scripts and shell commands to manage its data processing pipeline.\n
It calls a snapshot script from another skill: sh .agents/skills/meditate/scripts/snapshot.sh.\n
It uses a custom Python script: python3 .agents/skills/ruminate/scripts/extract-conversations.py.\n
It performs file system cleanup using rm -rf on the output directory.

ruminate