Skip to content

Spec: Velocity & Quality Roadmap (The "Engine First" Strategy)

Version: 2.0 — April 2026 Status: DRAFT Owner: Orchestrator Context: This roadmap pivots Agent Baton away from operational oversight and toward maximizing software development velocity and output quality. It focuses on the technical accelerators that turn the system into a compounding intelligence asset.

Wave 1: High-Throughput Parallelism

Goal: Enable multiple agents to work concurrently without friction or collisions.

1.1 Automated Git Worktree Isolation

  • Problem: Sequential execution is the primary bottleneck. Parallel execution in a single tree causes index locks and file collisions.
  • Requirement: Implement WorktreeManager to spin up isolated git worktrees for every parallel branch in a MachinePlan.
  • Velocity Impact: Enables 3-10× throughput by allowing true concurrent development across backend, frontend, and tests.

1.2 Declarative baton.yaml Workflow

  • Problem: Manual CLI configuration is overhead.
  • Requirement: A project-level config to define auto-routing, default quality gates, and agent preferences.
  • Velocity Impact: Reduces task setup time from minutes to seconds.

Wave 2: Compounding Institutional Memory

Goal: Formalize the processes for Knowledge Graph and Context Profile creation so the system gets smarter with every commit.

2.1 Formalized Knowledge Graph Synthesis

  • Problem: Beads are currently flat records. Relationships must be inferred manually.
  • Process Requirement: Implement a BeadSynthesizer that runs post-phase. It must:
    1. Infer Edges: Use semantic similarity and file-overlap to automatically link beads (e.g., "Feature A" extends "Architecture B").
    2. Cluster Discovery: Group related discoveries into "Knowledge Nodes."
    3. Conflict Detection: Automatically flag contradictory beads (e.g., two agents making different assumptions about an API).
  • Quality Impact: Prevents regression and ensures all agents work from a unified, evolving understanding of the codebase.

2.2 Formalized Agent Context Profile Harvesting

  • Problem: Agents start every task as "strangers" to the project.
  • Process Requirement: Implement a ContextHarvester that executes after every successful task_completed event.
    1. Expertise Extraction: Summarize the specific patterns and files the agent mastered during the task.
    2. Strategy Recording: Note which implementation strategies worked (passed gates) vs. which failed.
    3. Persistent Storage: Store in an agent_context table, indexed by (agent_name, domain).
  • Velocity Impact: Eliminates "cold start" token waste. Agents arrive at the next task already knowing the project's conventions and pitfalls.

Wave 3: Precision & Verification

Goal: Tighten the loop between planning, execution, and quality.

3.1 Expected Outcome (Demo Statements)

  • Problem: Passing tests != accomplishing the goal.
  • Requirement: Require an expected_outcome for every plan step. This becomes the primary prompt for the code-reviewer and test-engineer agents.
  • Quality Impact: Shifts the focus from "no errors" to "behavioral correctness."

3.2 Automated Handoff Synthesis

  • Problem: Context is lost during handoffs between agents.
  • Requirement: Auto-generate a "Handoff Bead" that summarizes the git diff, active discoveries, and remaining blockers from the previous agent.
  • Velocity Impact: Reduces the "re-orientation" phase for the next agent in the sequence.

Wave 4: Integration & Unblocking

Goal: Connect to external systems to automate the "last mile" of delivery.

4.1 CI-Driven Quality Gates

  • Problem: Local gates can't always replicate production environments.
  • Requirement: Direct integration with GitHub Actions/GitLab CI. The orchestrator dispatches the agent and then waits for the CI check to pass.
  • Quality Impact: Guarantees that agent-produced code meets the exact same standards as human PRs.

Removed/Deferred (Governance & Overhead)

The following items from the previous roadmap have been removed to focus on velocity: - PMO Dashboard: Deferred (secondary to engine performance). - Cost Accounting: Removed (operational overhead). - Incident Response Tooling: Removed (replaced by higher-quality verification). - Chargeback Systems: Removed.