The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses node to execute its management script council.mjs. It also relies on herdr to launch and control multiple model instances via herdr pane run, which involves executing CLI tools for various model families (e.g., claude, codex, agy).
[PERSISTENCE]: The included library file scripts/_lib/doctor.mjs contains functionality to modify the agent's local configuration file (.claude/settings.local.json). This is used to install hooks that trigger a hook-adapter.mjs script during session start and prompt submission, creating a persistent execution bridge for the framework's features.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest untrusted artifact files (such as external specifications or project plans) and provide them to multiple LLMs for review.
Ingestion points: The artifact file specified via the --artifact flag in council.mjs init.
Boundary markers: The artifact content is referenced in generated prompt files (round1.prompt.md), but the skill does not implement explicit sanitization or instruction-isolation delimiters around the artifact content itself.
Capability inventory: The agent ('chair') has the ability to write to the local filesystem and manage other model sessions via herdr.
Sanitization: No specific filtering or escaping is applied to the artifact content before it is read by the reviewing model seats.
[EXTERNAL_DOWNLOADS]: The providers.json metadata file lists installation commands for several third-party tools (e.g., jq, ripgrep, ast-grep, gitnexus) using standard package managers like brew, npm, and uv. These are well-known developer utilities.

council