tylergarlick/abraxas/pipeline-dispatcher

Pipeline Dispatcher Skill

Description

The Pipeline Dispatcher is the "Sovereign Reflex" of the OpenClaw environment. It automates the transition from task discovery (Beads) to task execution (Sub-agents), ensuring that no #ready task remains idle and that every worker is spawned with an appropriate complexity-based timeout.

Operational Logic

1. Discovery Phase

The dispatcher scans the bd (Beads) ledger for all open issues containing the #ready tag.

2. Estimation Phase (The Pre-Flight)

For each discovered task, the dispatcher interfaces with the pipeline-orchestrator MCP to:

• Deconstruct: Break the task into a concrete execution plan.
• Estimate: Assign a complexity score (Low, Medium, High, Critical).
• Time-box: Determine the timeoutSeconds based on the complexity (e.g., 300s for Low, 1800s for High).

3. Dispatch Phase

The dispatcher invokes sessions_spawn to create an isolated worker session.

• Input: The detailed plan from the orchestrator.
• Configuration: Set the runtime="subagent" and the calculated timeout.
• Tracking: Records the sessionKey of the spawned worker.

4. Provenance Recording

Every dispatch event is logged into the ArangoDB SovereignNodes and SovereignEdges collections:

• Node: Task $\rightarrow$ Edge: dispatched_to $\rightarrow$ Node: Session
• This creates a permanent, queryable record of exactly which agent handled which part of the project.

Commands

• /dispatch: Manually trigger a scan and spawn cycle.
• /dispatch-status: Report on all currently active workers and their linked beads.

Constraints

• No Double-Spawning: Tasks must be marked as in-progress immediately upon dispatch to prevent duplicate workers.
• Sovereign Guardrails: All spawned workers inherit the Sovereign prompt patch to ensure alignment.