The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell scripts to search for and read files on the host system, executing commands like ls, grep, jq, and cat with wildcards in the user's home directory (~/.clawdbot-* and ~/clawd-*) to find session logs and memory files.
[DATA_EXFILTRATION]: Accesses sensitive internal storage for agent conversation logs and shared memory files located at ~/.clawdbot-*/agents/*/sessions and ~/clawd-*/memory/. While used for context restoration, these files contain private conversation history and potentially sensitive data from past interactions.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection from untrusted data sources. Ingestion points: Retrieves external message history via message:read from chat platforms (Discord, Slack, etc.) and processes local .jsonl session files. Boundary markers: No explicit boundary markers or directives are used to distinguish untrusted recovered content from core instructions. Capability inventory: Includes shell command execution, file read access, and file-write access to memory directories. Sanitization: There is no evidence of input validation or sanitization of external content before it is integrated into the agent's working context.

context-recovery