Orchestrate a full feature lifecycle with human-in-the-loop gates across all SpecKit phases.

specify -> clarify -> plan -> checklist -> tasks -> analyze -> review -> implement -> verify -> CI

The fleet orchestrator chains 10 phases into a single command, detecting partially complete work and resuming from the right phase. It dispatches up to 3 parallel subagents during Plan and Implement, and uses a cross-model review before implementation to catch blind spots.

• 10-phase workflow -- end-to-end from idea to CI-passing code
• Human gates -- approve, revise, skip, abort, or rollback after every phase
• Mid-workflow resume -- detects existing artifacts and picks up where you left off
• Artifact integrity checks -- validates file sizes and expected sections during probe
• Stale artifact detection -- warns when upstream changes invalidate downstream files
• Parallel execution -- up to 3 concurrent subagents for [P]-marked tasks
• Cross-model review -- Phase 7 uses a different model to evaluate plan + tasks
• Implement-verify loop -- Phase 9 auto-remediates findings (up to 3 iterations)
• CI remediation loop -- Phase 10 auto-fixes test/build failures (up to 3 iterations)
• Phase rollback -- jump back to any earlier phase with downstream invalidation
• Branch safety -- pre-flight checks for uncommitted changes, detached HEAD, branch freshness
• Git checkpoints -- optional WIP commits after design, implementation, and verification
• Context budget awareness -- detects long sessions and suggests fresh-chat resume
• Completion summary -- structured report with artifact stats, quality gates, and git status
• IDE-agnostic -- works with VS Code Copilot, Claude Code, Cursor, and other platforms
• Verify integration -- auto-prompts to install the verify extension if missing

• Spec-Kit >= 0.1.0
• These core SpecKit commands must be available:
  • speckit.specify, speckit.clarify, speckit.plan, speckit.checklist
  • speckit.tasks, speckit.analyze, speckit.implement
• (Optional) Verify extension for Phase 9

specify extension add fleet --from https://github.com/sharathsatish/spec-kit-fleet/archive/refs/tags/v1.0.0.zip

specify extension add --dev /path/to/spec-kit-fleet

specify extension list
# Should show: fleet (1.0.0) -- Fleet Orchestrator

After installation, the following commands are registered:

In VS Code Copilot Chat:

/speckit.fleet Build a capability browser that lets users search and filter available capabilities

Or with no arguments to auto-detect progress on the current feature branch:

/speckit.fleet

The fleet automatically detects existing artifacts on every invocation. If you're on a feature branch with spec.md and plan.md already created, it will show:

Phase 1 Specify      [x] spec.md found
Phase 2 Clarify      [x] ## Clarifications present
Phase 3 Plan         [x] plan.md found
Phase 4 Checklist    [ ] --
...
> Resuming at Phase 4: Checklist

You can confirm or override to any phase.

/speckit.review

This runs the cross-model evaluation independently (read-only, no file modifications).

After installation, optionally customize settings by editing .specify/extensions/fleet/fleet-config.yml:

# Parallel execution (1-3 concurrent subagents)
parallel:
  max_concurrency: 3

# Model preferences -- set to exact model names for your IDE, or use defaults
# Examples:
#   VS Code Copilot:  "Claude Opus 4.6 (copilot)", "Claude Sonnet 4.6 (copilot)"
#   Claude Code:      "claude-sonnet-4-20250514", "claude-opus-4-20250514"
#   Cursor:           "claude-sonnet-4", "gpt-4o"
models:
  primary: "auto"     # Uses whatever model is running the fleet
  review: "ask"       # Prompts you on first run to pick a different model

# Verify extension auto-install prompt
verify:
  auto_prompt_install: true
  install_url: "https://github.com/ismaelJimenez/spec-kit-verify/archive/refs/tags/v1.0.0.zip"

On first run, the fleet detects your platform and asks which model to use for the cross-model review (Phase 7). The review should use a different model than the primary to catch blind spots -- e.g., if you're on Claude Opus, use GPT or Sonnet for review.

You can skip the prompt by setting explicit model names in the config file.

After every phase, you're asked to:

• Approve -- proceed to the next phase
• Revise -- re-run with feedback
• Skip -- skip this phase
• Abort -- stop the workflow

During Plan (Phase 3) and Implement (Phase 8), tasks marked with [P] are grouped into parallel batches:

<!-- parallel-group: 1 (max 3 concurrent) -->
- [ ] T002 [P] Create ModelA.cs
- [ ] T003 [P] Create ModelB.cs
- [ ] T004 [P] Create ModelC.cs

<!-- sequential -->
- [ ] T005 Create service that depends on all models

Constraints:

• Max 3 concurrent subagents per group
• Tasks touching the same file always run sequentially
• Human gate applies after each implementation phase completes

The fleet runs check-prerequisites.ps1 -Json -PathsOnly to discover FEATURE_DIR. This requires a feature branch with a matching specs directory. Ensure your branch follows the naming convention and a specs folder exists.

On first run, the fleet asks which model to use for review. To change it permanently, edit models.review in .specify/extensions/fleet/fleet-config.yml with your IDE's model identifier. Set to "ask" to be prompted again next time.

If Phase 9 reports the verify extension isn't installed, run:

specify extension add verify --from https://github.com/ismaelJimenez/spec-kit-verify/archive/refs/tags/v1.0.0.zip

Or set verify.auto_prompt_install: false in config to always skip verification.

This means an upstream file (e.g., spec.md) was modified after a downstream file (e.g., plan.md) was generated. Re-run the affected phase or proceed if the change was cosmetic.

1. Fork this repository
2. Create a feature branch
3. Make changes and test with specify extension add --dev /path/to/your-fork
4. Submit a pull request

MIT