Help us improve
Share bugs, ideas, or general feedback.
Experimental features for SPECTRE workflow framework.
npx claudepluginhub codename-inc/spectre-labsKnowledge capture and recall. Use /learn to save learnings, they'll be automatically recalled when relevant.
Share bugs, ideas, or general feedback.
Experimental and power-user features for the SPECTRE workflow framework.
Automated task execution CLI that runs Claude Code (or Codex) in a loop, completing one parent task per iteration with built-in code review and validation.
cd build-loop
pipx install -e .
pipx inject spectre-build pyyaml pydantic
# Interactive mode (prompts for inputs)
spectre-build
# Flag-based
spectre-build --tasks tasks.md --context scope.md --max-iterations 10
# With code review + validation (recommended)
spectre-build --tasks tasks.md --context scope.md --validate
# Plan mode — scope docs to build-ready manifest
spectre-build --plan --context scope.md design_notes.md
# Plan then auto-start build
spectre-build --plan --context scope.md --build
# Using a pipeline YAML definition
spectre-build --pipeline .spectre/pipelines/full-feature.yaml --tasks tasks.md
# From a manifest file (YAML frontmatter)
spectre-build docs/tasks/feature-x/build.md
# Resume interrupted session
spectre-build resume
# Start the web GUI
spectre-build serve
.md filesspectre-tdd skill for test-driven execution--plan transforms scope documents into a build-ready manifest. Instead of writing tasks by hand, provide scope/design docs and the pipeline produces a structured task breakdown you can feed directly into the build loop.
spectre-build --plan --context scope.md
The pipeline runs 6 stages autonomously:
Research → Assess → [Create Plan] → Create Tasks → Plan Review → Req Validate
|
PLAN_VALIDATED → build.md
CLARIFICATIONS_NEEDED → pause
Research — Explores the codebase, identifies key files, architecture patterns, and dependencies. Writes task_context.md.
Assess — Scores complexity and routes to the right depth:
LIGHT — Simple tasks skip plan generation, go straight to task breakdownSTANDARD — Normal planning depthCOMPREHENSIVE — Deep analysis with architecture diagrams and risk matricesCreate Plan — Writes an implementation plan with technical approach, critical files, and change boundaries. Skipped for LIGHT tasks.
Create Tasks — Breaks the plan into phased, ordered tasks with YAML frontmatter for the build manifest.
Plan Review — Catches over-engineering, unnecessary abstractions, and scope creep.
Req Validate — Cross-references the scope docs against the plan and tasks. Two outcomes:
PLAN_VALIDATED — Everything covered. Writes build.md manifest. Pipeline ends.CLARIFICATIONS_NEEDED — Gaps found. Writes a clarifications file, saves session, and pauses for human input.When the pipeline needs input, it pauses and notifies you:
📋 CLARIFICATIONS NEEDED
Edit: docs/tasks/main/clarifications/scope_clarifications.md
Then: spectre-build resume
Edit the file to answer the questions, then resume. The update_docs stage incorporates your answers and produces the final manifest.
After planning completes, start the build:
# Manual — review artifacts first, then run the manifest
spectre-build docs/tasks/main/build.md
# Automatic — chain directly with --build flag
spectre-build --plan --context scope.md --build
# Interactive — plan mode prompts "Start build now?" on completion
spectre-build
# → Mode [build/plan]: plan
Output artifacts land in docs/tasks/{branch}/:
docs/tasks/main/
├── build.md # Build manifest (YAML frontmatter)
├── task_context.md # Research findings
├── specs/
│ ├── plan.md # Implementation plan
│ └── tasks.md # Task breakdown
└── clarifications/
└── scope_clarifications.md # (if needed)
With --validate, spectre-build runs a 3-stage pipeline: