Skip to content

Process Document: bigpowers — Solo Developer SDLC

Owner: Solo Developer | Last Updated: 2026-06-13 | Version: 2.1.2 | Review Cadence: Per major release

Note: Version-specific sections tagged (v2.0.0 ...) describe features introduced in v2.0.0 that remain current unless noted.


Provide a deterministic, auditable, enterprise-ready software delivery lifecycle for solo developers. Eliminates the “what comes next?” problem through a conductor hierarchy, mandatory next-skill signaling, BCP scope accounting, and automated cycle-time metrics. Every story that ships writes its own evidence trail.


In scope: New features, bug fixes (fix-bug), refactors with a spec, projects of any size.

Out of scope: Emergency hotfixes (use release-branch --hotfix), pure research spikes (spike-prototype), multi-team orchestration (dispatch-agents), one-off document edits.


Step Responsible Accountable
Machine setup (seed-conventions) Developer Developer
Project scoping (orchestrate-project Ph 1–3) Agent + Developer Developer
Story planning (plan-work) Agent Developer
Branch + baseline (kickoff-branch) Agent Developer
TDD implementation (develop-tdd) Agent Developer
UAT sign-off (verify-work) Developer Developer
Code audit (audit-code) Agent Developer
Commit + land (commit-message + release-branch) Agent Developer
Metrics logging (cycle-times.yaml) Agent (automated) Developer
MVP release (semantic-release) Agent Developer

UAT (verify-work) is the only step where the Developer is exclusively Responsible. The agent cannot confirm behavioral correctness on behalf of the user.


ONE TIME (new machine or fresh clone)
────────────────────────────────────
seed-conventions
└─ Creates: CLAUDE.md, CONVENTIONS.md, .claude/, .gemini/, .cursor/rules/
└─ Syncs: skills via scripts/sync-skills.sh
└─ Gate: READY → next: orchestrate-project
ONCE PER PROJECT
────────────────
orchestrate-project --mode [standard | fast-track | ad-hoc]
├─ Phase 1 DISCOVER survey-context, research-first, elaborate-spec
├─ Phase 2 ELABORATE model-domain, grill-me, define-language, deepen-architecture, design-interface
├─ Phase 3 PLAN scope-work, slice-tasks, plan-work → release-plan.yaml (BCP baseline)
├─ Phase 4 BUILD → build-epic × N stories (see below)
├─ Phase 5 VERIFY run-evals, verify-work (project-level UAT)
└─ Phase 6 RELEASE commit-message, release-branch, semantic-release → v1.0.0 MVP
ONCE PER STORY (inside Phase 4 — called by build-epic)
───────────────────────────────────────────────────────
Step 1 survey-context ← stamps story_start timestamp
Step 2 plan-work ← writes [BCP N] tasks + verify: commands
Step 3 kickoff-branch ← creates worktree + feature branch
Step 4 develop-tdd ← RED → GREEN → REFACTOR (vertical slices)
Step 5 verify-work ← UAT gate (developer confirms behavioral correctness)
Step 6 audit-code ← quality gate ≥ 94%
Step 7 commit-message ← Conventional Commits + semver bump
Step 8 release-branch ← land to main; stamps story_end + writes cycle-times.yaml

See docs/references/workflow-steps.md for full per-step operational detail.


Scenario What to Do
Bug found mid-story Pause story (state.yaml preserves position). Run /fix-buginvestigate-bugdevelop-tddvalidate-fix. Resume at handoff.next_skill.
Gate fails (quality < 94%) Do NOT advance. Fix issues, re-run audit-code. Never lower threshold.
UAT step fails Return to develop-tdd for that BCP. Do not mark story done.
Session interrupted mid-story Run /survey-context — reads state.yaml epic_cycle and resumes at handoff.next_skill.
Task is ambiguous (no verify command) Do not write the task. Run define-success first.
Story blocked by external dependency Document in state.yaml handoff.open_decisions. Pause; work next WSJF story.
New requirement arrives mid-build Run change-request to insert into release-plan.yaml. Finish current story first.
sync-skills.sh fails on .gemini/ Sandbox-only issue. Run on host machine where .gemini/ is writable.
Need to parallel-ship two stories Use dispatch-agents inside develop-tdd. Never split the 8-step cycle across simultaneous stories on same worktree.

All gates must pass before advancing. Gates cannot be downgraded or skipped in Standard mode.

Gate type When Passes when Blocks on
Confirm plan→build, safety-critical actions Developer explicitly confirms No response or “no”
Quality audit-code npm run compliance ≥ 94% Score < 94%
Safety release-branch land to main Developer confirms “yes” No confirmation
Transition Phase advancement Required artifact files exist Missing state.yaml, SCOPE_LATEST.yaml, etc.
Hard Gate Inline in skills (plan-work, kickoff-branch) Specific precondition met Precondition not met — skill stops

Each critical-path skill writes state.yaml handoff.next_skill on exit. Run /survey-context at any time to read the current recommended next skill.

survey-context → plan-work
plan-work → kickoff-branch
kickoff-branch → develop-tdd
develop-tdd → verify-work
verify-work → audit-code
audit-code → commit-message
commit-message → release-branch
release-branch → survey-context (next story) | semantic-release (all done)