101 releases in 41 days. 33 epics delivered, 13 queued. 3 generations of software engineering thought, synthesized into executable agent discipline.

May 23 – June 10, 2026 · v1.0.0 → v1.x series

The launch. 38 spec-driven lifecycle skills covering the full arc: discover → elaborate → plan → build (TDD) → verify → release. Not a collection of tips — a methodology.

• Conventional Commits enforcement from day one. Every commit is feat/fix/docs, every merge triggers semantic-release. No manual version bumps.
• Agent workflow mandates baked into CLAUDE.md: agents cannot write code directly. Every task flows through a skill.
• Cursor + Gemini CLI + pi support generated automatically from a single source of truth (SKILL.md files).
• Karpathy behavioral mandates (write/select/compress/isolate) enforced from the start — token discipline isn't an afterthought, it's architecture.
• Clean Code heuristics, Superpowers gates, testing mandates — each hardened with Gherkin .feature files and npm run compliance scoring.

~25 releases in this phase. Each one a structural hardening, not a feature toss.

June 11–22 · v2.0.0 → v2.28.0

The major architecture change. Flat file structure couldn't scale — every new epic meant another loose markdown file in the root.

• v2.0.0 (June 11) — BREAKING. Capsule-directory architecture. Each epic gets its own directory with epic.yaml, numbered story specs, and a decoupled -tasks.yaml. ADR split (epic-local decisions stay with the epic, global decisions go to specs/adr/). Bug registry with structured naming (BUG-NNN-slug.md). Verification ledger. State lock file. 50/50 SDD adequacy score against a 22-method comparison.
• v2.10.0 (June 20) — CI verification and dry-run integrated into the skill system. Skills validate themselves.
• v2.20.0 (June 21) — Quality gates shipped. Audit-code gate, F.I.R.S.T. test rubric enforcement, universal checkpoint pattern across all workflow skills.
• v2.28.0 (June 22) — Machine-runnable skill catalog CI gate. Every SKILL.md's verify commands are now checked automatically on every push.

The system now builds itself. Skills create skills. Epics execute in 8 steps per story. The methodology is self-referential — bigpowers is built with bigpowers.

June 22–29 · v2.29.0 → v2.43.x

Production-grade infrastructure. The methodology was correct; now the pipeline had to be unbreakable.

• v2.29.0 (June 26) — Migrate-spec methodology complete. ID tracking, trace output, mandatory handoff, adversarial review. Absorbs learnings from migrating GSD, spec-kit, and BMAD projects.
• v2.30.0 (June 26) — Security-review integrated into all 9 workflow skills. Every PR gets a data-flow analysis before merge.
• v2.42.0 (June 29) — Standardized _LATEST.md naming. 7 skills updated. Predictable file names for CI pipelines.
• v2.44.0 (July 2) — Repository restructuring. 72 skill sources moved under skills/ directory. Root directory goes from ~96 entries to ~20. Clean landing page.

July 2–3 · v2.45.0 → v2.58.1

The quality infrastructure phase. This is where bigpowers proved it could measure and improve itself.

• G-04 sync-pipeline self-test: Every sync-skills.sh run validates its own output. Deterministic gate.
• Compliance CI gate: npm run compliance must pass ≥ 94% before any merge. Hard stop.
• Golden suite runner: Machine-runnable quality orchestration with baseline snapshots and size-change detection.
• Pre-merge mandate: No PR lands without compliance + sync + golden suite passing.
• Evolve-skill regression gate: Every skill change re-runs benchmarks to prove it's an improvement.

• Spike passed: DeepSeek v4 running via gh-aw Claude engine confirmed viable. G-01 golden story now lives in CI — a real agent runs a real skill on every push and proves it works.

• Git-derived effort: Hand-arithmetic cycle times replaced with commit-partition analysis. No self-reported wall-clock. Every number traceable to a git event.
• OKF provenance gate: Metrics bundles include aggregation tags and provenance pointers. Gate on the pipeline, not the value.

• Four strategies named and tooled: write (token-efficient code), select (only relevant context), compress (reduce without losing structure), isolate (partitioned worktrees).
• effort: heavy|light frontmatter on every skill. Agents can budget context before loading.

• Deterministic spec-to-code matrix: trace-stories.sh maps every story ID to implementing code. Three-tier confidence: tagged (high), heuristic (medium), inferred (low).
• Blind-spot detector: Heuristic analysis catches stories with no code match.
• CI/CD integration: Matrix runs on every push. 9 stories shipped in a single day — the fastest-moving epic in project history.

• Waiver subsystem: Deterministic scoring with explicit waiver exclusion. Every gate is auditable.
• G-07 negative-path self-test: Step scripts tested against deliberate failures. The golden suite now covers both "it works" and "it correctly fails."

The 13 gaps in the epic sequence are the roadmap. Every number below represents a queued epic, ordered by the project owner's priority.

• 101 releases in 41 days with zero manual version bumps. Semantic-release from Conventional Commits.
• 33 epics delivered through the same 8-step build cycle the methodology prescribes for users. bigpowers is its own first customer.
• Quality is machine-enforced, not human-asserted. Compliance scores, golden suites, sync-pipeline self-tests, traceability matrices, negative-path tests — all deterministic. No "trust me."
• Metrics are git-derived, not self-reported. Every cycle time, effort number, and lead time traces to a commit. No hand arithmetic.
• The stack is a chronological layer cake. Clean Code (2008) → Deep Modules (2018) → Agent Orchestration (2023) → Spec-Driven Development (2024) → AI-Native Code Hygiene (2026). Each wave resolves tensions from the last. Nothing is random.

"Simplicity is the ultimate sophistication, but integrity is the ultimate requirement."

verify: git tag --sort=-creatordate | wc -l should match the total releases count above (101 as of 2026-07-03).