The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reads historical conversation logs stored in JSONL format, which represent a source of untrusted data from previous sessions.
Ingestion points: Reads files from ~/.openclaw/agents/<agentId>/sessions/ using jq and rg.
Boundary markers: No specific delimiters or safety instructions are included to prevent the agent from following instructions found within the logs.
Capability inventory: The skill utilizes standard text processing tools (jq, rg, awk, sort). It does not use tools for network communication, file writing, or administrative tasks.
Sanitization: Uses jq to extract specific message fields, which offers structured data isolation but does not sanitize the text content for potential prompt injection attacks.
[COMMAND_EXECUTION]: The instructions provide several bash command templates that use pipes and subshells.
Evidence: Examples include complex for loops combined with jq, awk, and sort to generate summaries and cost reports from local files.

session-logs