SoloForge SOP — Internal Process Reference

This is a transparency tool. The agent follows this flow internally — the user doesn't need to memorize it. Print it when the user asks "where are we?" or wants to understand what the agent is doing.

Mark the current position with ➤ if you can infer it from context.

═══════════════════════════════════════════════════════════════
  PHASE 1: DISCOVER — Find what to build
═══════════════════════════════════════════════════════════════

Step 1.1 │ biz-think (Phase -1)
         │ You say: "幫我找產品機會"
         │ Agent: browses Reddit/Twitter/PH for pain signals
         │ Report: 3-5 opportunities with data + competitor analysis
         │ You: pick one / reject all
         │
         ↓
═══════════════════════════════════════════════════════════════
  PHASE 2: VALIDATE — Confirm it's worth building
═══════════════════════════════════════════════════════════════

Step 2.1 │ biz-think (Phase 0-5)
         │ You say: "評估這個方向"
         │ Agent: acid test, anti-patterns, revenue model, distribution
         │ Report: "Acid test [PASS/FAIL]. Recommend: [build/pivot/kill]"
         │ You: build → go to Phase 3
         │      pivot → go to Step 2.2
         │      kill  → back to Step 1.1
         │
         ↓ (if pivot or fail)

Step 2.2 │ biz-validate
         │ You say: "驗證假設"
         │ Agent: hypothesis → experiment → learn loop (max 5 rounds)
         │ Report: "Validation [PASS/FAIL]. Evidence: [summary]"
         │ You: validated → Phase 3
         │      kill → back to Step 1.1
         │
         ↓ (optional)

Step 2.3 │ brand-build
         │ You say: "幫我定品牌"
         │ Agent: positioning statement, voice, Build in Public strategy
         │ Report: "Positioning: [one sentence]. Voice: [3 adjectives]"
         │
         ↓
═══════════════════════════════════════════════════════════════
  PHASE 3: PLAN — Break it down
═══════════════════════════════════════════════════════════════

Step 3.1 │ project-init (new project only)
         │ You say: "初始化專案"
         │ Agent: git setup, CI/CD, staging env, CLAUDE.md
         │ Report: "Project [name] initialized. Git ✅ CI ✅ Staging ✅"
         │
         ↓

Step 3.2 │ work-breakdown
         │ You say: "拆解這個產品"
         │ Agent: read codebase → atomic issues → commit plans → test plans
         │ Report: "Milestone [name], [N] issues, test env needs: [list]"
         │ You: confirm / adjust issue list
         │
         ↓
═══════════════════════════════════════════════════════════════
  PHASE 4: BUILD — Code it (per issue, inner loop is autonomous)
═══════════════════════════════════════════════════════════════

You say: "做 Issue #N" or "build the milestone"
Agent runs the following automatically for each issue:

Step 4.1 │ issue-create
         │ Agent: create issue with AC, commit plan, test plan
         │ (skip if already created by work-breakdown)
         │
Step 4.2 │ pre-code
         │ Agent: issue check, branch check, scope boundaries, platform config
         │
Step 4.3 │ [write code]
         │ Agent: implement according to commit plan
         │
Step 4.4 │ self-review
         │ Agent: read diff, correctness, design, security, scope, output verification
         │
Step 4.5 │ test-gate
         │ Agent: verify unit + integration + E2E tests exist and pass
         │
Step 4.6 │ pre-commit
         │ Agent: atomic commit check, conventional format, no debug/secrets
         │
Step 4.7 │ pre-pr
         │ Agent: PR size check, links issue, targets dev
         │ If UI changes → ux-audit before merge
         │
Step 4.8 │ [merge to dev]
         │
Step 4.9 │ post-merge (partial)
         │ Agent: verify fix works, check AC boxes, close issue, delete branch
         │ Report: "Issue #N done. [N/total] complete. Changes: [summary]"
         │
         │ Repeat 4.1-4.9 for each issue until milestone complete
         │
         ↓
═══════════════════════════════════════════════════════════════
  PHASE 5: VERIFY — Deploy + test
═══════════════════════════════════════════════════════════════

Step 5.1 │ staging-verify
         │ You say: "準備發布"
         │ Agent: delta analysis, env vars check, platform config, structural check
         │ Report: "Structural check [PASS/FAIL]. Config: [complete/gaps]"
         │ If fail → fix gaps → re-run Step 5.1
         │
         ↓

Step 5.2 │ [merge dev → main]
         │ Agent: create PR, merge to main, wait for deploy
         │
         ↓

Step 5.3 │ post-merge (production testing)
         │ Agent: Playwright tests on production URL (full user journey)
         │ Report: "Production test: [N] bugs found. Please test [URL]."
         │
         ↓

Step 5.4 │ [You test]
         │ You: test the product yourself at production URL
         │ You: report bugs / feedback
         │ Agent: creates issues for all bugs → back to Phase 4 if needed
         │
         ↓ (when you're satisfied)

Step 5.5 │ product-eval (optional)
         │ You say: "評估產品品質"
         │ Agent: Nielsen heuristics, cognitive walkthrough, FTUE, HEART
         │ Report: "Score: [N]/100. Grade: [letter]. Top issues: [list]"
         │
         ↓
═══════════════════════════════════════════════════════════════
  PHASE 6: SHIP — Release + retrospective
═══════════════════════════════════════════════════════════════

Step 6.1 │ post-merge (release)
         │ You say: "發布"
         │ Agent: create GitHub release with semantic versioning
         │ Report: "v[X.Y.Z] released. [link]"
         │
         ↓

Step 6.2 │ retro
         │ Agent: analyze git metrics, identify improvements
         │ Report: "Retro: [metrics]. Action items: [list]"
         │
         ↓

Step 6.3 │ roadmap-steer
         │ You say: "下一步做什麼"
         │ Agent: RICE scoring, prioritize next feature
         │ Report: "Next feature: [name]. RICE: [score]"
         │ You: approve → back to Step 3.2
         │
         ↓ (cycle repeats)

═══════════════════════════════════════════════════════════════
  AVAILABLE ANYTIME (you can invoke these at any point)
═══════════════════════════════════════════════════════════════

  "幫我寫文案"     → copy-craft        "幫我寫內容"    → content-create
  "幫我看定價"     → finance-ops       "幫我看法律"    → legal-guard
  "幫我看品牌"     → brand-build       "幫我做社群"    → community-engage
  "幫我看數據"     → data-decide       "幫我做銷售"    → sales-close
  "幫我看基礎設施"  → infra-ops        "幫我看客服"    → support-ops
  "幫我發佈內容"   → content-distribute "幫我看增長"    → growth-track
  "幫我排優先級"   → roadmap-steer     "幫我看搜尋結果" → search-eval
  "流程圖" / "/sop" → this flowchart

Next Steps

After printing, tell the user where you think they currently are (if you can infer it from context), and suggest what they might want to do next.