The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill is configured to index and access files explicitly described as containing authentication data. In SKILL.md, the team collection targets ~/.openclaw/agents/**/*.json, which is documented to store 'auth' and agent configurations.
[DATA_EXFILTRATION]: The skill exposes sensitive system and personal information by indexing private directories. It indexes the ~/.openclaw/ directory for system configurations and ~/clawd/memory/ for daily work logs, historical decisions, and project status.
[COMMAND_EXECUTION]: The script memory_router.sh executes a CLI tool using a user-provided query variable. The $QUERY variable is passed directly to the qmd query command; without sanitization, this creates a potential command injection surface if the query contains shell metacharacters or if the qmd tool handles arguments unsafely.

memory-router