Claude Code plugin for conceptualizing, designing, implementing, updating, and validating complex Mistral Vibe workflows.
VibeFlow is for workflows where a generic prompt is not enough: multi-phase automation, custom skills, tools, MCP/connectors, middleware, hooks, agent profiles, subagents, source-level changes, and validation evidence. Its job is to preserve design freedom while forcing every selected runtime surface to be grounded in what Mistral Vibe can actually do.
VibeFlow separates two questions that often get mixed together:
- What is the best workflow design for this task?
- Which Mistral Vibe runtime surfaces can honestly implement that design?
The plugin should not tell the agent to use every possible customization point. It should choose the smallest sufficient surface, explain why it fits, reject unnecessary surfaces with rationale, and record enough evidence that implementation oversight is visible later.
- init — interactive intent loop. Completes only after user sign-off and generation of
VISION.md,PLAN.md, andWORKFLOW_CONTRACT.json. - design — maps signed intent onto real Mistral Vibe runtime surfaces. Compares grounded candidate architectures, lints the candidate set, then promotes the winning candidate into component breakdowns, diagrams, feasibility classification, design decisions, rejected alternatives, and approval-ready artifacts.
- plan — researches source/docs as needed and produces implementation targets, component contracts, tests, and validation gates.
- apply — writes or patches files according to the approved plan. Runs a pre-apply surface guard to enforce the contract before any file is touched.
- validate — runs the serial validation chain, writes evidence, classifies failures, detects drift, and reports whether the workflow is ready or needs rework.
- update — modifies or hardens an existing workflow through narrative intake, repo/artifact scan, ambiguity clarification, proposed edits, approval, implementation, record updates, and validation.
- inspect — audits an existing workflow/repo and persists an inspection report for later design/update work.
- realize — ingests existing conceptual workflows, skillsets, plugin drafts, partially implemented repos, or design documents and converts them into working, runtime-grounded implementations.
VibeFlow's references model the current Mistral Vibe runtime surfaces. Key points:
- Skills:
allowed_toolsis advisory only. Use agent profileenabled_toolsfor actual restriction. - Middleware: A loop guard, not a phase orchestrator.
before_turn()fires before every LLM call in the tool loop (N+1 times per multi-tool turn). Registration requires source modification (Tier D). - Hooks:
POST_AGENT_TURNonly. Exit 2 + non-empty stdout = retry. 3-retry ceiling per hook per turn. - Agent profiles: Per-phase tool enforcement via
enabled_tools/disabled_tools. Profile switching is NOT subagent isolation. - Compaction: Agent profile survives natively.
todostate does not. Persist important state to disk.
Full reference: references/feasibility/runtime-pattern-catalog.md
Agent runtime model: references/feasibility/agent-runtime-model.md
Implementation patterns: references/feasibility/implementation-patterns.md
Key interfaces: references/feasibility/key-interfaces-contracts.md
Every phase produces a machine-checkable artifact set that the next phase consumes without re-interpreting intent.
| Phase | Consumes | Produces |
|---|---|---|
| init | User interaction | VISION.md, PLAN.md, WORKFLOW_CONTRACT.json |
| design | WORKFLOW_CONTRACT.json, references/feasibility/* |
DESIGN_CANDIDATES.json, DESIGN.md, ARCHITECTURE.md, diagram |
| plan | WORKFLOW_CONTRACT.json, DESIGN.md, ARCHITECTURE.md |
Updated PLAN.md with file targets, contracts, validation gates |
| apply | PLAN.md, DESIGN.md, ARCHITECTURE.md, WORKFLOW_CONTRACT.json |
Implementation files, workflow manifest |
| validate | Repo state, manifest, WORKFLOW_CONTRACT.json |
Validation report, evidence, drift classification |
| realize | Existing artifacts, references/feasibility/* |
REALIZATION_CONTRACT.json, implementation patches, report |
WORKFLOW_CONTRACT.json is the durable contract between phases. It records signed intent, selected/rejected runtime surfaces, rationale, capabilities, contracts, evidence targets, and change-control records.
The schema does not force every workflow to include every surface. It forces the workflow to justify the surfaces it selects and explain why others are unnecessary.
Schema: references/manifest-schema.json
applymust fail if it implements a surface not selected in the contract.applymust fail if it omits a selected surface without a recorded amendment.validatemust classify drift (missing surface, unauthorized surface, wrong surface, impossible assumption, stale evidence, validation gap).updateis the only phase allowed to amend the contract after initial sign-off.
See references/examples/ for workflow examples and the archetype catalog. Start with the archetype guide (references/examples/archetypes/) to identify which workflow shape fits your task before looking at any specific example.
Load the plugin directly:
claude --plugin-dir /path/to/VibeFlowpython3 /path/to/VibeFlow/scripts/install-to-claude-profiles.py
python3 /path/to/VibeFlow/scripts/install-to-claude-profiles.py --apply --install-cacheVibeFlow uses two layers of instructions:
commands/— User-facing slash command definitions. Each command file has YAML frontmatter (name,description) that Claude Code uses to decide when to offer the command. The command body is a brief pointer to the backing skill. These files are not the detailed instructions; they are the entry point.skills/— Detailed behavioral instructions loaded by Claude Code when a command is invoked. Each skill is aSKILL.mdwith YAML frontmatter (name,description,allowed-tools,user-invocable) plus the full step-by-step procedure the agent follows.
The command's description field is the contextual trigger — it tells Claude when this command is relevant. The skill's description provides additional matching context. Keep descriptions specific to avoid false triggers:
| Command | Trigger context | Backing skill |
|---|---|---|
/vibe-workflow-init |
Starting a new Mistral Vibe workflow from scratch | skills/vibe-workflow-init/SKILL.md |
/vibe-workflow-design |
Mapping workflow intent to runtime topology | skills/vibe-workflow-design/SKILL.md |
/vibe-workflow-plan |
Creating implementation targets from a design | skills/vibe-workflow-plan/SKILL.md |
/vibe-workflow-apply |
Writing/patching files from an approved plan | skills/vibe-workflow-apply/SKILL.md |
/vibe-workflow-validate |
Proving a workflow works with evidence | skills/vibe-workflow-validate/SKILL.md |
/vibe-workflow-update |
Modifying or hardening an existing workflow | skills/vibe-workflow-update/SKILL.md |
/vibe-workflow-inspect |
Auditing an existing repo or workflow | skills/vibe-workflow-inspect/SKILL.md |
/vibe-workflow-realize |
Converting a concept/partial into a working workflow | skills/vibe-workflow-realize/SKILL.md |
Trigger design principle: Descriptions should mention "Mistral Vibe" and "workflow" to avoid firing on unrelated tasks. They should not mention generic terms like "design" or "plan" alone, which would trigger on non-VibeFlow work.
Example sequence:
/vibe-workflow-init
/vibe-workflow-design
/vibe-workflow-plan
/vibe-workflow-apply
/vibe-workflow-validate
PYTHONPYCACHEPREFIX=/tmp/vibeflow-pycache python3 -m py_compile scripts/*.py
python3 -m json.tool references/manifest-schema.json
PYTHONPYCACHEPREFIX=/tmp/vibeflow-pycache python3 -m unittest discover -s tests -vcommands/— user-facing slash command definitions.skills/— command backing instructions loaded by Claude Code.references/— shared Mistral Vibe runtime and feasibility knowledge.references/feasibility/— runtime contracts, pattern catalog, implementation patterns, key interfaces, and Mermaid diagrams.references/examples/— workflow examples and archetype catalog.scripts/— validators, linters, simulators, drift detection, contract enforcement, install/sync tooling.tests/— regression tests for workflow tooling and plugin guarantees.
VibeFlow is for advanced Mistral Vibe users building multi-phase workflow commands, quality gates, custom tools, MCP integrations, middleware-backed behavior, and source-level customizations.
It is not intended as a beginner Mistral AI guide.
VibeFlow is still hardening. The current focus is validation integrity, runtime-surface accuracy, update/inspect loops, prevention of plausible but non-functional designs, and pattern-fit enforcement.