The SPEED + Claude Code Workflow

This workflow combines the orchestration power of SPEED, the ad-hoc agility of Claude Code, and the workspace persistence of TMUX. We’ll use the bookshelf example to illustrate.

Phase 1: Context & Discovery (Claude Code)

Before triggering a formal SPEED run, use Claude Code CLI in your main TMUX window to probe the project.

• Objective: Understand where a new feature fits.
• Action: claude "find where the book search logic is and explain the schema"
• Why: Claude Code is much faster for “read-only” research and map-making than spawning full agent worktrees.

Phase 1.5: The Structural Audit (SPEED)

Before moving to planning, audit your new spec to ensure it’s structurally sound and appropriately sized.

• Action: speed audit specs/tech/my-feature.md
• Result: SPEED checks your spec against the official template and warns you if the feature is too big (sizing warning). If a spec is too large, split it now to avoid context overload later.

Phase 2: The Pre-Plan Workflow (SPEED)

Once you have an initial direction, run the formal SPEED pre-plan battery to ensure your specifications are high-fidelity and vision-aligned.

1. Validate: speed validate specs/ - Catches orphans, contradictions, and missing links across your product, tech, and design specs.
2. Guard: speed plan <spec> - This automatically triggers the Pre-Plan Guardian Gate, which uses the Product Guardian agent to check if your technical proposal actually aligns with the overview.md vision.
3. Verify: speed verify - An adversarial “blind check” where an agent verifies the task DAG against the spec without seeing the Architect’s reasoning.

Phase 3: Monitoring the Fleet (TMUX)

This is where the Ghostty-style TMUX setup shines. Your screen becomes a command center:

• Window 1 (Main): Your primary editor (Neovim/Helix) stays in the bookshelf root.
• Window 2 (Monitor): speed status --verbose
  • You see Agent-1 (Developer) implemented the search endpoint in a worktree.
  • You see Agent-2 (Reviewer) catches an off-by-one error in the pagination.
• Window 3 (Claude Help): Keep a Claude Code instance ready in a side pane for quick “How do I…” questions during implementation if you need to manually intervene.

Phase 4: Quality Gates & Integration

As tasks complete, SPEED’s Quality Gates run within their isolated worktrees.

• Failure Analysis: If a test fails, the Failure Analyst agent reports it to your TMUX log window.
• Merge: Once all gates are green, speed integrate merges the worktree branches back to main.

Summary Layout

Using the Ghostty-style splits:

+-----------------------------------------------------------+
| [SPEED] [bookshelf] [L1:Index]                            |
+-----------------------------+-----------------------------+
|                             |                             |
|          EDITOR             |       SPEED LOGS            |
|       (Neovim/Helix)        |    (speed status)           |
|                             |                             |
+-----------------------------+-----------------------------+
|        CLAUDE CODE          |       SHELL / TESTS         |
|      (Ad-hoc probing)       |      (Manual Check)         |
+-----------------------------+-----------------------------+