Strategic-to-Tactical Workflow Skill

III. The Complete Workflow

This is an 8-phase workflow from strategic tension to tactical execution.

Phase 1: Recognize the Tension

Goal: Identify and articulate the strategic tension or question that needs exploration.

Pattern: The best work begins not with a command ("build X"), but with a tension ("should we do X or Y?").

Actions:

1. Listen for the tension in the user's request
2. Articulate it as a clear, open-ended question
3. Resist the urge to immediately solve it

Example (v0.0.31):

• Tension: "Should Dojo Genesis be a feature lab (25 pages) or a focused desktop product (single-page)?"
• Not: "Build a desktop app"

Output: A clearly articulated strategic tension.

Related Skills: strategic-scout

Phase 2: Scout Multiple Routes

Goal: Explore 3-5 distinct strategic routes before committing to any one path.

Pattern: Scouting prevents waste. 30 minutes of route exploration can save weeks of misguided work.

Actions:

1. Use /strategic-scout to generate 3-5 routes
2. For each route, define: approach, risks, impact, duration
3. Identify tradeoffs and dependencies
4. Present routes without recommending one initially

Example (v0.0.31):

• Route 1: Big Bang (rebuild everything)
• Route 2: Incremental (gradual migration)
• Route 3: Parallel Tracks (split into independent workstreams)
• Route 4: Hybrid (mix of approaches)
• Route 5: Minimal (smallest possible change)

Output: A set of viable routes with clear tradeoffs.

Related Skills: strategic-scout, iterative-scouting

Phase 3: Gather Feedback & Listen for Reframes

Goal: Present the routes and listen for the question behind the question.

Pattern: The first scout is for provocation, not consensus. The goal is to elicit a deeper, more insightful framing.

Actions:

1. Present the scouted routes
2. Listen not just for agreement/disagreement, but for how the feedback is framed
3. Identify if the user is introducing a new lens or reframe
4. If a reframe emerges, prepare to re-scout

Example (v0.0.31):

• Initial framing: "Deprecate web app vs. keep as companion"
• Reframe: "Desktop for deep work vs. Mobile for on-the-go orchestration"
• Result: Re-scout with the new lens (led to mobile PWA as premium tier)

Output: Either a decision on a route, or a reframe that triggers a second scout.

Related Skills: iterative-scouting, product-positioning

Phase 4: Make the Strategic Decision

Goal: Commit to a strategic direction based on the scouting and feedback.

Pattern: Decisions unlock velocity. Once the route is chosen, move quickly to specification.

Actions:

1. Select the best route (or hybrid of routes)
2. Articulate the decision clearly
3. Document the why behind the decision
4. Identify what this decision enables (and what it rules out)

Example (v0.0.31):

• Decision: Parallel Tracks approach (Route 3)
• Why: Balances speed (parallelization) with risk management (clear boundaries)
• Enables: 3 agents working simultaneously
• Rules out: Big Bang (too risky), Incremental (too slow)

Output: A clear strategic decision with rationale.

Related Skills: strategic-scout

Phase 5: Ground in the Codebase

Goal: Before writing specifications, deeply understand the current state of the codebase.

Pattern: Grounding beats assumptions. Specs based on outdated or assumed knowledge lead to rework.

Actions:

1. Use /repo-context-sync to pull latest code and understand structure
2. Read relevant files (backend APIs, frontend components, architecture docs)
3. Document backend architecture, API endpoints, data models
4. Identify integration points and dependencies
5. Create a backend grounding document

Example (v0.0.31):

• Read go_backend/main.go, handlers/, middleware/
• Document all API endpoints relevant to desktop UI
• Create v0_0_31_backend_grounding.md
• Identify SSE streaming architecture for orchestration events

Output: A comprehensive backend grounding document.

Related Skills: repo-context-sync, frontend-from-backend

Phase 6: Write Specifications (With Parallel Track Decomposition)

Goal: Transform the strategic decision into detailed, implementation-ready specifications.

Pattern: Specifications enable parallelization. Good specs are force multipliers.

Actions:

1. Decide if the work should be split into parallel tracks
2. If yes, apply /parallel-tracks:
   • Identify natural boundaries (by layer, feature, component)
   • Define track dependencies and execution order
   • Create dependency graph
3. For each track (or single spec), write:
   • Architecture document: High-level design, component structure, data flow
   • Specification document: Detailed requirements, success criteria, non-goals
4. Validate using /parallel-tracks-validation checklist

Example (v0.0.31):

• Track 1: Router Migration & Shell (foundation, no dependencies)
• Track 2: Sidebar 1 (depends on Track 1)
• Track 3: Sidebar 2 (depends on Track 1)
• Track 4: Main Content Area (depends on Track 1)
• Execution: Track 1 first, then Tracks 2-4 in parallel

Output: Complete specifications for each track, validated and ready for commission.

Related Skills: frontend-from-backend, parallel-tracks, release-specification

Phase 7: Write Implementation Prompts & Commission

Goal: Transform specifications into structured implementation prompts and commission the work.

Pattern: Prompts are commissions. The quality of the prompt determines the quality of the implementation.

Actions:

1. For each track/spec, write an implementation prompt using the standard template:
   • Context & Grounding: Link to spec, pattern files, files to read/modify
   • Detailed Requirements: Step-by-step, numbered, specific
   • File Manifest: Files to create/modify
   • Success Criteria: Binary, testable checkboxes
   • Constraints & Non-Goals: Explicit boundaries
   • Backend Grounding: API contracts, integration points
2. Validate each prompt against the quality checklist
3. Commission prompts to appropriate agents:
   • Zenflow: Strategic implementation (architecture, complex features)
   • Claude Code: Tactical implementation (UI components, refactoring)
4. If parallel tracks, commission in dependency order

Example (v0.0.31):

• Write v0_0_31_track_1_implementation_prompt.md
• Commission Track 1 to Claude Code
• Wait for Track 1 completion
• Commission Tracks 2-4 to Claude Code in parallel

Output: Commissioned work in progress, with clear success criteria.

Related Skills: write-implementation-prompt, implementation-prompt

Phase 8: Reflect & Create Meta-Skills

Goal: After execution, reflect on what worked, extract patterns, and formalize them as skills.

Pattern: Every cycle produces wisdom. The goal is not just to ship software, but to improve how we ship software.

Actions:

1. Perform /compression-ritual to preserve key insights
2. Identify patterns that emerged during the cycle:
   • Strategic patterns (how we thought, decided, navigated uncertainty)
   • Tactical patterns (how we specified, commissioned, integrated)
   • Meta-patterns (how we learned, iterated, improved)
3. Use /process-extraction to formalize valuable patterns as skills
4. Use /seed-extraction to capture smaller insights as seeds
5. Commit all artifacts (specs, prompts, compressions, skills) to repository

Example (v0.0.31):

• Created product-positioning skill (strategic pattern)
• Created multi-surface-strategy skill (strategic pattern)
• Created iterative-scouting skill (meta-pattern)
• Created parallel-tracks seed (tactical pattern)

Output: New skills and seeds added to the knowledge base, ready for the next cycle.

Related Skills: compression-ritual, process-extraction, seed-extraction

IV. The Loop Closes

After Phase 8, the loop closes. The next development cycle begins with Phase 1, but now we have:

• More skills to guide the process
• More patterns to recognize and apply
• More wisdom about what works and what doesn't

This is the essence of continuous improvement. Each cycle makes the next cycle faster, clearer, and more effective.

V. Key Principles

1. Start with Tension, Not Solutions

The best work begins with a well-articulated tension, not a predetermined solution. Hold the question open long enough for better answers to emerge.

2. Scout Before Committing

Always explore multiple routes before choosing one. The first idea is rarely the best idea.

3. Ground in Reality

Specifications based on assumptions fail. Always ground in the current state of the codebase.

4. Parallelize When Possible

With good specifications and clear boundaries, multiple agents can work simultaneously without conflicts.

5. Formalize What Works

Every successful cycle should produce new skills and patterns. This is how we compound our capabilities.

VI. Example: v0.0.31 Complete Workflow

Total planning time: ~6 hours
Implementation time: 6-8 days (with parallelization)
Outcome: v0.0.31 shipped, 3 new skills created, workflow improved

VII. Quality Checklist

Before moving to the next phase, ensure you can answer "yes" to these questions:

Phase 1:

• Have you articulated the tension as an open-ended question?
• Have you resisted the urge to immediately solve it?

Phase 2:

• Have you scouted at least 3 distinct routes?
• Have you identified risks, impact, and duration for each route?

Phase 3:

• Have you listened for reframes in the user's feedback?
• Have you decided whether to commit or re-scout?

Phase 4:

• Have you made a clear strategic decision?
• Have you documented the "why" behind the decision?

Phase 5:

• Have you pulled the latest code and read relevant files?
• Have you created a backend grounding document?

Phase 6:

• Have you decided whether to parallelize the work?
• Have you written complete specifications for each track?
• Have you validated the parallel track structure?

Phase 7:

• Have you written structured implementation prompts for each track?
• Have you commissioned the work in the correct dependency order?

Phase 8:

• Have you compressed the context and extracted patterns?
• Have you created new skills or seeds from the learnings?
• Have you committed all artifacts to the repository?

VIII. Related Skills

This skill is the meta-pattern that connects all other strategic and tactical skills:

Strategic Skills:

• strategic-scout - Route exploration
• product-positioning - Reframing binary decisions
• multi-surface-strategy - Complementary product surfaces
• iterative-scouting - Scout → feedback → reframe → re-scout

Tactical Skills:

• repo-context-sync - Codebase grounding
• frontend-from-backend - Backend-first specification
• parallel-tracks - Parallel track decomposition
• write-implementation-prompt - Structured prompt writing
• implementation-prompt - Spec-to-prompt transformation

Meta-Skills:

• compression-ritual - Preserve insights
• process-extraction - Formalize patterns
• seed-extraction - Capture smaller learnings

Output

• Phase 1-2: Scout document (thinking/[topic]_strategic_scout.md) with tension, routes, and selected direction
• Phase 5: Backend grounding document (docs/vX.X.X/[release]_backend_grounding.md) mapping API endpoints, data models, and integration points
• Phase 6: Track specification documents (docs/vX.X.X/[release]_track_[N]_spec.md), one per parallel workstream
• Phase 7: Implementation prompt files (docs/vX.X.X/prompts/track_[N]_prompt.md), self-contained and commission-ready
• Phase 8: New skills or seeds in the knowledge base, and a compression note summarizing key learnings from the cycle

Examples

Scenario 1: "We're starting v0.0.32. We're not sure whether to prioritize the API layer or the frontend shell." → Full 8-phase run: tension named ("API completeness vs. UI polish"), 5 routes scouted, decision made (parallel tracks: API track + UI shell track), codebase grounded, two spec docs written, two implementation prompts commissioned, post-cycle reflection extracts one new pattern as a seed.

Scenario 2: "We shipped v0.0.31. What did we learn and what should we formalize?" → Phase 8 only: compression ritual surfaces 3 patterns from the cycle, one is strong enough to become a skill (written and committed), two become seeds, all artifacts committed to the repository.

Edge Cases

• If the user is mid-cycle (e.g., specs are written but prompts are not), enter at the appropriate phase rather than restarting from Phase 1 — ask which phase they are on before proceeding.
• If the codebase has not been read recently (Phase 5 skipped in a prior cycle), treat grounding as mandatory even if the user wants to move quickly — specs built on stale assumptions produce rework.
• For small changes (a single-file bug fix or a minor UI tweak), this 8-phase workflow is overkill — redirect to strategic-scout for just the decision, then directly to an implementation prompt. Reserve the full workflow for releases that span multiple files or architectural decisions.

Anti-Patterns

• Running Phase 6 (spec writing) before Phase 5 (codebase grounding) — specifications written without reading the current code routinely miss existing abstractions, causing implementation agents to rebuild what already exists.
• Treating Phase 8 as optional — skipping reflection means the patterns from this cycle are not available to the next cycle, which defeats the compounding-velocity goal of the workflow.
• Commissioning all parallel tracks simultaneously before Track 0 (the foundation track) is verified — always complete the dependency-order gate before releasing parallel agents.
• Using this skill as a checklist to tick boxes rather than a thinking tool — the value is in the quality of the tension articulation and the depth of the codebase grounding, not in completing all 8 phases mechanically.

IX. The Vision

This workflow is not static. It will evolve with each cycle. The goal is to reach a state where:

1. Strategic decisions are fast and confident (because we have patterns to recognize and skills to apply)
2. Specifications are grounded and actionable (because we have systematic grounding processes)
3. Implementation is parallel and autonomous (because we have clear boundaries and comprehensive prompts)
4. Learning is continuous and formalized (because we extract and document patterns after every cycle)

This is the path to compounding velocity: each cycle makes the next cycle faster, clearer, and more effective.

This is the Dojo way.

🪷