run

Autodialectics Run

Use this skill to drive the Autodialectics anti-slop pipeline against a task.

Prerequisites

autodialectics-mcp must be on PATH (pip install autodialectics).

MCP Workflow

1. health — verify the MCP server is reachable.
2. init_runtime — ensure the database and default champion policy exist.
3. compile_task — compile the task JSON into an immutable contract. Inspect it before proceeding.
4. run_task — execute the full pipeline (evidence → dialectic → execution → verification → evaluation → gate).
   • Pass detach: true for long-running tasks. Poll with inspect_run afterward.

CLI Fallback

If the MCP server is not available:

autodialectics init
autodialectics compile <task.json>
autodialectics run <task.json>

Task File Format

Tasks are JSON files with at minimum:

{
  "title": "Short description",
  "description": "What needs to be done and why",
  "domain": "code|research|writing|experiment|analysis|generic",
  "workspace_root": "path/to/workspace",
  "verification_commands": ["pytest -q"],
  "assets": [{"kind": "file", "location": "path/to/file", "label": "name"}]
}

Guidance

• Always compile before running if the task is new or ambiguous.
• Summaries should include: run ID, decision (accept/reject/revise/rollback), overall score, slop composite, and unresolved risks.
• Pass a policy_id to run_task to test a specific policy instead of the current champion.

Arguments

If the user passes a task path after /autodialectics:run, treat it as the target task file and execute the compile → run workflow.