The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reads project requirements from an external file (brief.md) located in the workspace. If this file contains malicious instructions, it could attempt to influence the agent's behavior during the configuration generation process.
Ingestion points: Reads content from /opt/yeticlaw/openclaw/workspace/projects/[slug]/brief.md via the cat command.
Boundary markers: None identified. The agent is instructed to read the brief directly into its context.
Capability inventory: The skill has the ability to list directories (ls), create directories (mkdir), copy files (cp), and write files (cat >) within the /opt/yeticlaw/openclaw/workspace/ path.
Sanitization: The skill is constrained to output only valid JSON, which mitigates the risk of the injection directly controlling the output format, but not the content.
[COMMAND_EXECUTION]: The skill utilizes several shell commands to manage the game development environment.
Evidence: Uses ls, cat, mkdir, and cp to interact with files and directories within the /opt/yeticlaw/openclaw/workspace/ directory structure. These commands are used to discover templates, read schemas, and deploy the HTML engine template to project folders.

threejs-dev