The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements an automated pipeline that promotes content from untrusted session logs into persistent system instructions.
Ingestion points: The skill reads JSONL transcript files located in ~/.openclaw/agents/main/sessions/ via the summarize-sessions.sh script.
Boundary markers: None are implemented. The instructions do not provide guidance on how to distinguish legitimate user feedback from malicious prompt injection payloads contained within the logs.
Capability inventory: The skill possesses the ability to modify critical workspace files including AGENTS.md, TOOLS.md, and other SKILL.md files, which govern the agent's behavior and available tools.
Sanitization: No sanitization, filtering, or human-in-the-loop validation is performed before insights are extracted and written to the configuration files.
[COMMAND_EXECUTION]: The skill utilizes shell commands and local scripts to aggregate session data.
Executes openclaw sessions to identify target log files.
Uses tail and an inline Python processor within scripts/summarize-sessions.sh to parse and filter chat history.

self-reflection