The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the subprocess.run function across several core scripts to manage its evolutionary lifecycle. Evidence includes:
scripts/apply-proposal.py executes Git commands (git init, git config, git add, git commit) to establish version control and provide rollback capabilities for workspace modifications.
scripts/orchestrator.py invokes the platform's agent binary (openclaw agent --local) to process data worksets for reasoning and analysis.
scripts/workspace-init.py runs openclaw --version to verify environment compatibility.
[INDIRECT_PROMPT_INJECTION]: The skill implements a data ingestion surface by scanning conversation history and recording user feedback, which could contain malicious instructions.
Ingestion points: scripts/scan-history.py extracts messages from historical session logs, and scripts/signal-record.py stores raw user feedback in a local database.
Boundary markers: No specific delimiters or "ignore previous instructions" guards are implemented in the automated scanning logic.
Capability inventory: The skill can modify core configuration files (SOUL.md, AGENTS.md) and execute shell commands via Git and the platform agent.
Sanitization: The risk is mitigated by a mandatory human-in-the-loop requirement where all proposed changes must be explicitly reviewed and accepted by the user before any file-write or execution occurs.

byted-ark-evolve