Flow

A cross-tool agentic-coding workflow toolkit: 12 workflow skills (research → plan → implement → validate → review → ship), an optional cross-vendor code-review skill, and a tracker abstraction (GitHub Issues, Azure DevOps work items, GitLab issues, …) - all behind the Agent Skills open standard so the same files work in Claude Code and GitHub Copilot CLI.

Install

Claude Code (plugin, recommended)

/plugin marketplace add corticalstack/flow
/plugin install flow@flow
/flow:init

/flow:init is an interactive bootstrap that creates AGENTS.md at your project root from Flow's template, creates .claude/tracker.json from the example, and optionally sets up cross-vendor review profiles.

GitHub Copilot CLI

gh skill install corticalstack/flow                                  # interactive picker (recommended)
gh skill install corticalstack/flow .claude/skills/create-plan       # one specific skill

Then drop in an AGENTS.md (see docs/AGENTS.md.template) and configure .claude/tracker.json (see .claude/tracker.example.json).

Cloned repo (template-style, alternative)

If you want the entire repo - including the optional Ralph autonomous-loop scripts under scripts/ and worked examples:

gh repo create my-project --template corticalstack/flow

See docs/TEMPLATE_INSTRUCTIONS.md for post-clone setup.

What you get

Workflow skills (12):

• /flow:research-requirements, /flow:research-codebase - structured research phase
• /flow:create-plan, /flow:iterate-plan - implementation planning
• /flow:implement-plan, /flow:validate-plan - execution + quality gate
• /flow:commit, /flow:autonomous-commit - commit hygiene
• /flow:describe-pr, /flow:handle-pr-feedback - PR description + review-feedback loop
• /flow:create-handoff, /flow:resume-handoff - cross-session continuity

Optional cross-vendor review:

• /flow:cross-review - a configurable OAuth-only review against Azure AI Foundry, GitHub Models, or GitHub Copilot CLI. A model from a different vendor reads the diff vs main and surfaces high/critical findings - advisory only, never edits code. See docs/cross-vendor-review.md.

Tracker abstraction:

• bin/tracker (PATH-wrapper) + .claude/scripts/tracker.sh (the adapter)
• github backend wired (uses gh); azure-devops backend stub; none for tracker-less projects
• Configurable per-project via .claude/tracker.json. See docs/tracker-portability.md.

Bootstrap helper:

• /flow:init - creates AGENTS.md, .claude/tracker.json, and optional cross-review/profiles.json from templates.

Workflow overview

/flow:research-requirements
       ↓
/flow:create-plan
       ↓
/flow:implement-plan        ←─── iterate per phase
       ↓
/flow:validate-plan          (quality gate)
       ↓
/flow:cross-review           (optional, advisory)
       ↓
/flow:commit
       ↓
Push  &  PR
       ↓
/flow:describe-pr
       ↓
Review (human or @claude)
       ↓
/flow:handle-pr-feedback     (if changes requested)
       ↓
Merge

Each stage drops a durable markdown artifact under flow/research/, flow/plans/, flow/prs/, or flow/reviews/, so the next agent (or future you) can pick up cleanly.

Docs

• docs/tracker-portability.md - the tracker adapter contract, neutral state vocabulary, supported backends.
• docs/cross-vendor-review.md - /flow:cross-review setup, backends, caveats.
• docs/AGENTS.md.template - the AGENTS.md /flow:init ships.
• docs/TEMPLATE_INSTRUCTIONS.md - for the cloned-repo path.

License

MIT - see LICENSE.

Contributing

See CONTRIBUTING.md.