# Intentgraph

> Spec-driven graph of intent, constraints, decisions, and rationale. Provides MCP tools for querying, mutating, and verifying the IntentGraph for the current workspace.

- Skill: `quinnaho/intentgraph` (Agent Skill, multi-file: 7 files)
- Install (CLI): `npx skillmds@latest add quinnaho/intentgraph`
- Raw SKILL.md: https://api.skillmd.com/api/skills/quinnaho/intentgraph/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: quinnaho (https://skillmd.com/u/quinnaho)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/quinnaho/intentgraph

---


# IntentGraph Skill

Activate this skill when the user asks about intents, constraints, decisions, drift, or wants to propose or verify a code change. The skill connects to the IntentGraph subprocess for the current workspace and exposes MCP tools listed above.

Default flow:

1. `retrieval.search` to find relevant intent/constraint/concept nodes for the user's request.
2. `graph.get_node` to expand context for the top hits.
3. If proposing a change: `task.create` → `task.lease` → make edits in a shadow worktree → `task.propose_patch` → preview, then `task.accept_patch` or `task.reject_patch`.
4. `verify.run` against affected obligations before accepting any patch.
5. Always cite the node IDs you traversed in your final response — the retrieval IS the externalized reasoning.