The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple shell commands for orchestration, including git operations for source control, python3 for prompt rendering and logic, and jq for data processing. It also dynamically resolves and executes repository formatters like biome, ruff, black, and prettier based on the detected project environment.\n- [DATA_EXFILTRATION]: Performs network operations via an integrated api tool to communicate with endpoints such as /task, /projects, and /activity. These calls are used to synchronize task state, fetch project configuration, and record agent activity logs as part of the intended workflow.\n- [EXTERNAL_DOWNLOADS]: Resolves model configurations from local environment variables and models.json, and may trigger package managers (npm, yarn, pnpm, bun) to execute formatting scripts defined within the user's project.\n- [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface. Ingestion points include user-controlled task fields (title, description) and agent-generated comments. These are interpolated into subagent templates using bracketed placeholders (e.g., ). The system's capabilities include shell execution and API writes, while sanitization is managed through structured parsing (jq/python) and internal filtering principles.

squad-run