The Agent Skills Directory

[DATA_EXFILTRATION]: The skill explicitly accesses sensitive agent-specific directories, including the memory path (e.g., ~/.claude/projects/<project>/memory/). While intended for auditing interaction history, these files contain records of past sessions and potentially sensitive context that is exposed to the subagents.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes untrusted content from the 'brain/' vault and project memory. Maliciously crafted content within these processed files could influence the behavior of the Auditor and Reviewer subagents.
Ingestion points: Reads all markdown files in the brain/ directory and logs in the agent's memory folder.
Boundary markers: The skill lacks explicit instructions for subagents to distinguish between structural metadata (like file headers) and potentially adversarial content within the files being audited.
Capability inventory: The skill has the capability to rewrite or delete files in the knowledge base, agent instructions (e.g., AGENTS.md, CLAUDE.md), and memory files based on subagent findings.
Sanitization: No sanitization or filtering is performed on the ingested data before it is passed to the subagents for analysis.
[COMMAND_EXECUTION]: The skill executes a local shell script scripts/snapshot.sh to perform file aggregation. While the script itself is a standard utility using find and cat, it facilitates the automated reading of large portions of the filesystem into temporary snapshots.

meditate