The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes external CLI tools (such as claude, codex, or gemini) as defined in the council.config.yaml configuration file. This is implemented using Node.js child_process.spawn without a shell, which effectively prevents command injection via user-provided prompts. The implementation logic is found in scripts/council-job-worker.js where the command string is parsed into a program and an arguments array.
[EXTERNAL_DOWNLOADS]: The scripts/council-job.js script includes a recommendation to install the yaml dependency or update the skill via npx github:team-attention/agent-council if requirements are not met. This targets the author's own official repository for project-specific tooling and is a standard deployment method for this utility.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a proxy, ingestion user prompts and passing them directly to multiple configured agent CLIs for opinion gathering.
Ingestion points: The user's prompt is captured in scripts/council-job.js (line 330) and passed to the worker processes.
Boundary markers: None are explicitly used to wrap or delimit the user input before it is passed to the sub-agents.
Capability inventory: The skill possesses the capability to spawn arbitrary processes as defined in its local configuration file.
Sanitization: No content-based sanitization is performed on the user's prompt, although the skill mitigates command injection by avoiding shell execution in the spawn call in scripts/council-job-worker.js (line 114).

agent-council