The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/check.py utility uses subprocess.run to execute local Python scripts for skill validation and auditing. These scripts include a bundled asset toolkit and may also attempt to execute scripts from parent directory structures if they exist.
[EXTERNAL_DOWNLOADS]: The README provides installation instructions using npx, which fetches and executes code from a remote registry. It also identifies its upstream provenance from a GitHub repository.
[REMOTE_CODE_EXECUTION]: The skill instructs the agent to utilize tools for spawning autonomous sub-agents (team_spawn) and merging their results (team_merge). While these are standard capabilities for the coordination framework, they represent high-autonomy operations that are gated by recommended review processes.
[PROMPT_INJECTION]: The coordination workflow creates a surface for indirect prompt injection where a lead agent ingests data from sub-agents. Ingestion points: The lead agent reads results and updates via team_results and team_message. Boundary markers: The instructions mandate gates for diff inspection and result verification. Capability inventory: The session uses team_spawn, team_merge, and shell command execution. Sanitization: The skill relies on the lead agent's reasoning to inspect integrated changes rather than automated input filtering.

opencode-ensemble