iteration: @mellanon — JellyBean: Spec Driven Development

[mellanon]

Champion: @mellanon
Period: Feb 3 – Feb 14, 2026
Goal: SpecFlow lifecycle direction established, brownfield capability designed and spec'd

Vision

Extend SpecFlow from a build tool (SPECIFY→COMPLETE) into a full lifecycle tool covering review, release, and brownfield evolution. The council verdict (C+) says build these capabilities natively in SpecFlow — don't depend on OpenSpec as a runtime dependency, don't wrap in Maestro playbooks. Extract the delta-spec pattern from OpenSpec, port the 8-gate release framework from SpecFirst, own the implementation.

The research foundation is committed. This iteration translates that research into concrete SpecFlow feature specs that can be implemented.

In Flight

• 🏃 Upstream PRs landed — headless pipeline and bug fixes merged to jcfischer/specflow-bundle (PRs signal: Package and submit PR to danielmiessler/PAI #3, specflow-lifecycle: Align with @jcfischer on extending SpecFlow bundle #4, specflow-lifecycle: Design native specflow review command #6, specflow-lifecycle: Design native specflow release command (port SpecFirst 8-gate) #7)
• 🏃 Brownfield capability — designed and spec'd, ready for implementation in specflow-bundle

Planned

• Review command — specflow review designed, informed by four-layer review strategy (specflow-lifecycle: Design native specflow review command #6)
• Release command — SpecFirst 8-gate framework ported to SpecFlow design (specflow-lifecycle: Design native specflow release command (port SpecFirst 8-gate) #7)
• 💪 Cedars engagement — respond to @Steffen025's milestone-based approach (governance: SpecFirst - Milestone-based alternative to phase-based development #72)

Completed

• Research foundation — landscape report, OpenSpec deep dive, council debate (see References)
• Project direction pivoted — native SpecFlow commands, not Maestro playbooks
• Issues re-scoped — specflow-lifecycle: Native contrib-prep process documentation #5 (contrib-prep docs), specflow-lifecycle: Design native specflow review command #6 (review command), specflow-lifecycle: Design native specflow release command (port SpecFirst 8-gate) #7 (release command) updated per council verdict
• Governance updated — research/ directory, CONTRIBUTING.md, onboarding SOP

Deferred

• ⬛ OpenSpec interchange format — evaluate at 90-day review gate, not now (specflow-lifecycle: Draft Open Spec template #8)

References

Where the work happens:

• Implementation repo: mellanon/specflow-bundle (fork — spoke repo where features are built)
• Upstream: jcfischer/specflow-bundle (PRs target here)
• Project registration: projects/specflow-lifecycle/PROJECT.yaml

Research (foundation for this iteration):

• Spec-Driven Dev Landscape — ecosystem map, tool comparison, architecture
• OpenSpec Deep Dive — delta-spec semantics, risk assessment
• Council Debate: SpecFlow vs OpenSpec — verdict C+

Prior art:

• SpecFirst source: contrib-specfirst-v1.0.0 branch (8-gate release framework to port)

[jcfischer]

Upstream PR status — @jcfischer

Merged two PRs to jcfischer/specflow-bundle today:

PR	Title	Status
#3	Fix migrations not running in compiled binary	Merged
#4	Allow N/A sections in verify.md for CLI-only features	Merged (council-reviewed)
#6	AI-powered headless Doctorow Gate	Changes requested — council review posted
#7	Full headless mode for autonomous SpecFlow pipelines	Changes requested — council review posted

Council findings on #6 and #7

Both PRs are solid architecturally but have blocking issues:

PR #6 (headless Doctorow Gate):

• Fail-open default on a quality gate undermines its purpose — should be fail-closed with opt-in fail-open
• Process zombie leak in timeout handler

PR #7 (headless pipeline):

• Headless mode omits --dangerously-skip-permissions — needs verification that file writes work
• Default model Opus costs $5-8/pipeline run — Sonnet recommended ($1.50-2.50)
• process.env mutation should be replaced with explicit options parameter

Full reviews with code examples are on the upstream PRs.

[jcfischer]

Council Verdict: SpecFlow Direction Decided (C+)

The council debate on SpecFlow vs OpenSpec has concluded with a near-unanimous verdict. Full debate: research/2026-02-02-council-debate-specflow-vs-openspec.md

Verdict: Option C+ — Build Natively, Emit Compatibly

4/4 council members agreed on:

• Delta-spec concept (ADDED/MODIFIED/REMOVED) has clear value for brownfield workflows
• SpecFlow's quality gates, TDD enforcement, and Doctorow Gate are non-negotiable — must be preserved
• OpenSpec as a direct dependency is too risky (bus factor=1, supply chain concerns)
• Build brownfield capability natively in SpecFlow

Recommended path:

1. Build specflow brownfield with delta-spec semantics on SQLite state
2. Port SpecFirst 8-gate release framework + CHANGELOG management into SpecFlow
3. Optionally emit OpenSpec-compatible format as interchange — but don't depend on it
4. Set 90-day review gate to assess whether interchange format adds value
5. Maintain Claude-only trust boundary; reject 23-tool integration bloat

Impact on Specflow-Lifecycle Issues

Issue	Impact
#5 (contrib-prep docs)	✅ Confirmed — native process, no external deps
#6 (review command)	✅ Confirmed — build natively
#7 (release command)	✅ Confirmed — port SpecFirst 8-gate framework
#8 (Open Spec template)	⚠️ Reframed — interchange format only, not a dependency. Lower priority pending 90-day review

Remaining Disagreements

• Format output: Should SpecFlow emit OpenSpec-compatible interchange? (Ava: yes; Marcus/Rook: unnecessary coupling)
• Timeline: Marcus estimates 2-3 days for brownfield; Rook says audit extraction difficulty first
• Scope: Whether this is "workflow tool vs platform" or "own your infrastructure"

What This Means for This Iteration

The strategic direction is locked. The specflow-lifecycle work (#5-8) proceeds as planned with native implementations. The council verdict validates the current issue set — no pivot required, just confirmation that the build-natively approach is correct.

The Cedars collaboration proposal from @Steffen025 (#72) remains complementary — Cedars as session orchestrator, SpecFlow as the spec engine inside each session.