The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local shell script located at scripts/snapshot.sh to aggregate markdown files from the brain/ and .agents/skills/ directories. While the script performs standard file concatenation, it represents a point of local command execution.
[INDIRECT_PROMPT_INJECTION]: The skill processes extensive amounts of data from the brain/ directory and ~/.claude/ memory directories which may contain untrusted or third-party content. Malicious instructions embedded in these files could potentially influence the sub-agents (Auditor and Reviewer) during the analysis phase.
Ingestion points: Content is ingested from brain/, .agents/skills/, CLAUDE.md, and the agent's internal memory directory ~/.claude/projects/*/memory/ via the snapshot.sh script.
Boundary markers: The snapshots use === path === headers as file delimiters, but the instructions provided to the sub-agents in references/agents.md do not include explicit 'ignore embedded instructions' warnings or safety boundaries for the data being parsed.
Capability inventory: The skill has the capability to execute shell scripts (sh), read/write files throughout the project vault, and delete files (notes and memories) during the 'Apply changes' step.
Sanitization: There is no evidence of sanitization, filtering, or instruction-extraction prevention applied to the data before it is passed to the sub-agents for analysis.
[DATA_EXPOSURE]: The skill consolidates the entire knowledge base and skill set into single, unencrypted temporary files at /tmp/brain-snapshot.md and /tmp/skills-snapshot.md. This concentrates potentially sensitive project information and internal agent configurations into easily accessible temporary locations.

meditate