fleet

Fleet Orchestrator Extension

All-in-one lifecycle orchestrator for Spec Kit. Drives a feature from idea to merged code through 17 phases autonomously — with bundled Definition of Done, Codebase Impact, UX Research, Stitch prototyping, and Code Quality sub-commands. One specify extension add fleet brings the full pipeline.

Inspired by good ideas from the community: circuit breaker (Ralph), progress.md log (Ralph), machine-readable status tracker (Product Forge), sync-verify (Product Forge), change-request (Product Forge), post-implementation quality pipelines, adversarial multi-model code review (Anvil pattern), and explicit --skip-* bypass flags (plan-review-gate).

Phases

#	Phase	Optional	Command	Artifact
1	Specify	—	speckit.specify	spec.md
2	DoD Generate	--skip-dod	speckit.fleet.dod-generate	dod.yml
3	Clarify	--skip-clarify	speckit.clarify	## Clarifications in spec.md
4	Plan	—	speckit.plan	plan.md
5	Codebase Impact	--skip-impact	speckit.fleet.codebase-impact-analyze	codebase-impact.md
6	UX Research	--skip-ux (auto-skip if no UI)	speckit.fleet.ux-research-analyze	ux-research-report.md
7	Checklist	--skip-checklist	speckit.checklist	checklists/
8	Tasks	—	speckit.tasks	tasks.md
9	Analyze	—	speckit.analyze	.analyze-done marker
10	Review	--skip-review	speckit.fleet.review	review.md (cross-model)
11	Stitch Prototype	--skip-stitch (auto-skip if no UI)	speckit.fleet.stitch-prototype	.stitch-prototype-done
12	Implement	—	speckit.implement	all [x] in tasks.md
13	DoD Validate	--skip-dod	speckit.fleet.dod-validate	.dod-validate-done
14	Stitch Validate	--skip-stitch (auto-skip if no UI)	speckit.fleet.stitch-validate	.stitch-validate-done
15	Code Review	--skip-code-review	speckit.fleet.quality-pipeline + adversarial	reviews/quality-summary.md + adversarial-review.md
16	Release Readiness	--skip-release	fleet orchestrator	release-readiness.md
17	Tests	—	terminal	CI passes

Key behaviors:

• Resumes from the correct phase automatically — run speckit.fleet.run on any branch, at any point
• Autonomous by default — uses vscode_askQuestions only for critical blockers (FAIL/CRITICAL findings, circuit breaker, missing extensions) and final ship approval (Phase 16)
• WIP auto-commits after every artifact-producing phase (wip(fleet): phase {N} {name}), controlled by git.auto_commit config
• Auto-stashes uncommitted changes before the run (git stash push -m "fleet-auto-stash: ...") and reminds at completion, controlled by git.auto_stash config
• Auto-skips Phase 10 when models.review is "ask" (unconfigured) — no prompting on first run
• CI auto-fix — first iteration auto-fixes without asking; iteration 2+ uses vscode_askQuestions
• Parallel subagents (up to 3) during Plan and Implement for [P]-marked tasks
• Circuit breaker: 3 consecutive zero-progress implement batches → halt and ask
• progress.md log: timestamped entry after every completed phase or explicit skip/override decision, enables fast resume across sessions
• .fleet-status.yml tracker: machine-readable phase state, powers the sync command
• Context budget management with compact summaries between phases
• Phase 2 DoD Generate produces machine-readable dod.yml from spec acceptance criteria; Phase 13 DoD Validate checks implementation against it
• Phase 5 Codebase Impact scans for integration points and produces IMPACT-NNN task candidates that feed into tasks.md
• Phase 10 uses a different model than the rest of the workflow to catch blind spots
• Phase 15 Code Review runs speckit.fleet.quality-pipeline then an adversarial multi-model pass — 2-3 review subagents on different AI models with consensus scoring
• Phase 16 Release Readiness generates a READY / CONDITIONAL / NOT READY checklist
• Phases 6, 11, and 14 auto-skip when the feature has no UI (keyword detection in spec.md/plan.md)

Install

specify extension add fleet --from https://github.com/danieldekay/spec-kit-extensions/archive/refs/heads/main.zip

This single install brings all 20 commands — orchestration, DoD, codebase impact, UX research, Stitch prototyping, and code quality. No companion extensions required.

VS Code Copilot users

After installing the extension, run this once to copy the agent files to .github/agents/:

/speckit.fleet.agents-install

Or copy manually:

mkdir -p .github/agents
cp .specify/extensions/fleet/agents/speckit.fleet.*.agent.md .github/agents/

Commands

Orchestration

Command	Alias	Description
speckit.fleet.run	speckit.fleet.go	Start or resume the full fleet workflow
speckit.fleet.review	—	Cross-model review of design artifacts (invoked by fleet automatically)
speckit.fleet.agents-install	—	Install VS Code Copilot agent files to .github/agents/
speckit.fleet.sync	—	Cross-cutting artifact drift detector (7 consistency layers)
speckit.fleet.change-request	—	Formal scope change with CR-NNN tracking and artifact markers

Definition of Done (Phases 2 & 13)

Command	Alias	Description
speckit.fleet.dod-generate	speckit.fleet.dod-gen	Parse spec.md and generate dod.yml with testable DoD criteria
speckit.fleet.dod-validate	speckit.fleet.dod-check	Validate implementation against dod.yml and update statuses
speckit.fleet.dod-export	speckit.fleet.dod-sf	Export dod.yml to specfact-compatible JSON for CI/CD
speckit.fleet.dod-report	speckit.fleet.dod-rpt	Generate human-readable DoD status report

Codebase Impact (Phase 5)

Command	Alias	Description
speckit.fleet.codebase-impact-analyze	speckit.fleet.impact	Scan codebase for integration points and produce IMPACT-NNN candidates

UX Research (Phase 6)

Command	Alias	Description
speckit.fleet.ux-research-analyze	speckit.fleet.ux	Analyze spec for UX needs and discover existing patterns

Stitch MCP (Phases 11 & 14)

Command	Alias	Description
speckit.fleet.stitch-prototype	speckit.fleet.stitch-ui	Generate UI screens via Stitch MCP
speckit.fleet.stitch-validate	—	Validate implemented UI against Stitch prototypes
speckit.fleet.stitch-setup	—	Discover Stitch projects and design masters, write config override

Code Quality (Phase 15)

Command	Alias	Description
speckit.fleet.quality-pipeline	speckit.fleet.qa	Full pipeline: review → fix → validate → future ideas → specfact sync
speckit.fleet.code-review	speckit.fleet.cr	Code review: refactoring, tech debt, dead code, smells
speckit.fleet.code-fix	speckit.fleet.cf	Auto-fix issues from the code review
speckit.fleet.validate-requirements	speckit.fleet.vr	Validate FR/NFR implementation, testability, and docs
speckit.fleet.future-ideas	speckit.fleet.fi	Generate ideas for improving and extending the feature
speckit.fleet.specfact-sync	speckit.fleet.sf	Export quality findings to specfact-compatible JSON

Configuration

After installing, generate a config file:

cp .specify/extensions/fleet/config-template.yml .specify/extensions/fleet/fleet-config.yml

For personal overrides (gitignored — never commit model choices):

touch .specify/extensions/fleet/fleet-config.local.yml
echo '.specify/extensions/fleet/fleet-config.local.yml' >> .gitignore

Config precedence (highest wins): CLI flags > env vars > fleet-config.local.yml > fleet-config.yml > extension defaults.

Config sections

The config file has 6 namespaced sections. See config-template.yml for all options with comments.

Section	Governs	Key settings
fleet	Orchestrator	models.primary, models.review, phases.skip_*, qa.*, git.*, adversarial.*
dod	Definition of Done	criteria_per_fr, nfr_defaults, enforcement_level, specfact.*
codebase_impact	Codebase Impact	scan_depth, scan_directories, exclude_directories, risk_threshold
ux_research	UX Research	scan_depth, pattern_directories, tech_stack, component_registries
stitch	Stitch MCP	mode, variant_count, projects[], design_masters[]
code_quality	Code Quality	review.*, fix.*, validate.*, future.*, specfact.*

Key fleet settings

Setting	Default	Description
fleet.models.primary	"auto"	Model for most phases
fleet.models.review	"ask"	Model for Phase 10 — auto-skips when "ask", set a model name to enable
fleet.phases.skip_*	false	Skip individual phases
fleet.qa.cadence	"per_phase"	When to run quality pipeline: per_phase or batch_end
fleet.qa.pause_on_critical	true	Pause Fleet if CRITICAL findings remain
fleet.adversarial.enabled	true	Enable multi-model adversarial review in Phase 15
fleet.adversarial.models	["gpt-4.1", "gemini-2.5-pro", "claude-sonnet-4"]	Models for adversarial review
fleet.git.auto_commit	true	WIP auto-commit after every artifact-producing phase
fleet.git.auto_stash	true	Auto-stash uncommitted changes before fleet run

Schemas

File	Description
schemas/dod.schema.json	JSON Schema for dod.yml — validates DoD structure
schemas/specfact-export.schema.json	JSON Schema for specfact export format (format ID: speckit-dod-export-v1)

Requires

• speckit >= 0.2.0
• Core SpecKit commands: speckit.specify, speckit.clarify, speckit.plan, speckit.checklist, speckit.tasks, speckit.analyze, speckit.implement
• Optional: Stitch MCP server >=1.0.0 for Phases 11/14

Workflow Diagram

Fleet wraps the entire Spec Kit core flow and interleaves extension phases where they add the most value. Core commands run unconditionally (or with simple skip flags); bundled phases degrade gracefully when auto-skip conditions are met.

flowchart TD
    classDef core fill:#1f6feb,stroke:#1f6feb,color:#fff
    classDef bundled fill:#8957e5,stroke:#8957e5,color:#fff
    classDef fleet fill:#da3633,stroke:#da3633,color:#fff

    PRE["🔒 Pre-flight<br/>git stash · HEAD check · branch freshness<br/>artifact detection · resume logic"]

    P1["① Specify<br/><b>speckit.specify</b><br/>→ spec.md"]:::core
    P2["② DoD Generate ◇<br/><b>speckit.fleet.dod-generate</b><br/>→ dod.yml"]:::bundled
    P3["③ Clarify ◇<br/><b>speckit.clarify</b><br/>→ §Clarifications"]:::core
    P4["④ Plan<br/><b>speckit.plan</b><br/>→ plan.md"]:::core
    P5["⑤ Codebase Impact ◇<br/><b>speckit.fleet.codebase-impact-analyze</b><br/>→ codebase-impact.md"]:::bundled
    P6["⑥ UX Research ◇<br/><b>speckit.fleet.ux-research-analyze</b><br/>→ ux-research-report.md"]:::bundled
    P7["⑦ Checklist ◇<br/><b>speckit.checklist</b><br/>→ checklists/"]:::core
    P8["⑧ Tasks<br/><b>speckit.tasks</b><br/>→ tasks.md"]:::core
    P9["⑨ Analyze<br/><b>speckit.analyze</b><br/>→ .analyze-done"]:::core
    P10["⑩ Review ◇<br/><b>speckit.fleet.review</b><br/>→ review.md"]:::bundled
    P11["⑪ Stitch Prototype ◇<br/><b>speckit.fleet.stitch-prototype</b><br/>→ .stitch-prototype-done"]:::bundled
    P12["⑫ Implement<br/><b>speckit.implement</b><br/>→ all ☑ in tasks.md"]:::core
    P13["⑬ DoD Validate ◇<br/><b>speckit.fleet.dod-validate</b><br/>→ .dod-validate-done"]:::bundled
    P14["⑭ Stitch Validate ◇<br/><b>speckit.fleet.stitch-validate</b><br/>→ .stitch-validate-done"]:::bundled
    P15["⑮ Code Review ◇<br/><b>speckit.fleet.quality-pipeline</b><br/>+ adversarial multi-model<br/>→ reviews/ + adversarial-review.md"]:::bundled
    P16["⑯ Release Readiness ◇<br/>fleet orchestrator<br/>→ release-readiness.md"]:::fleet
    P17["⑰ Tests<br/>terminal CI runner<br/>→ tests pass"]:::fleet

    PRE --> P1
    P1 --> P2 --> P3 --> P4 --> P5 --> P6
    P6 --> P7 --> P8 --> P9 --> P10 --> P11
    P11 --> P12 --> P13 --> P14 --> P15
    P15 --> P16 --> P17

Loading

Legend: ◇ = optional / skippable  |  🟦 core spec-kit  |  🟪 bundled  |  🟥 fleet-only

Migration from v0.x (separate extensions)

If you previously had fleet, dod, codebase-impact, ux-research, stitch-implement, and code-quality installed as separate extensions:

1. Remove the old extensions: specify extension remove dod codebase-impact ux-research stitch-implement code-quality
2. Update fleet: specify extension update fleet
3. Update command references in any custom scripts: speckit.dod.* → speckit.fleet.dod-*, speckit.codebase-impact.* → speckit.fleet.codebase-impact-*, etc.