The Agent Skills Directory

[COMMAND_EXECUTION]: The skill communicates with a local daemon service via curl at http://localhost:4632. These commands are used to initiate sub-agents, monitor their status, and retrieve final task results.- [COMMAND_EXECUTION]: It invokes a local Bun script located at ~/.claude/plugins/cache/claudeclaw/claudeclaw/1.0.0/commands/progress.ts to provide visual feedback on the progress of parallel tasks.- [DATA_EXFILTRATION]: No data exfiltration to external or untrusted domains was detected. All network communication is directed to the loopback interface (localhost).- [INDIRECT_PROMPT_INJECTION]: The skill aggregates outputs generated by sub-agents, creating an ingestion surface for potentially untrusted data. 1. Ingestion points: Reads sub-agent results from local files in ~/.claude/claudeclaw/subagents/ and from the REST API responses. 2. Boundary markers: No explicit delimiters or instructions to ignore embedded commands are used when assembling the final report. 3. Capability inventory: The skill can execute local scripts via bun, interact with local APIs via curl, and perform file system operations (read/write/list) within the specific application data directory. 4. Sanitization: No content validation or sanitization is performed on the data received from sub-agents before it is merged into the final output.

subagent-runner