Instructions file
Claude Code instructions for mtrdesign/whygraph, covering claude.md, status, common commands, documentation site and before pushing.
Instructions file
Claude Code instructions for mtrdesign/whygraph, covering claude.md, status, common commands, documentation site and before pushing.
Agent
WhyGraph plan executor. Reads a reviewed plan markdown, applies each step in order, runs the per-step Verify, and logs progress back into the plan. Spawned by the /whygraph-implement slash command — do not invoke for unrelated work.
Agent
WhyGraph planning orchestrator. Given a code-change task, builds a working set of affected symbols via CodeGraph, warms WhyGraph rationale cache, then either writes a plan (single-pass mode) or fans out to 3 researchers + a synthesizer (deep mode). Spawned by the /whygraph-plan slash command — do not invoke for…
Agent
WhyGraph fan-out researcher. Investigates one specific dimension of a planned code change (impact, constraintsrisks, or priorart) using WhyGraph + CodeGraph MCP tools. Spawned in parallel with sibling researchers by the whygraph-planner in deep mode — do not invoke directly.
Agent
WhyGraph fan-in synthesizer. Combines reports from 3 fan-out researchers (impact, constraintsrisks, priorart) into a single step-by-step implementation plan in WhyGraph's standard format. Spawned by the whygraph-planner in deep mode after researchers complete — do not invoke directly.
Skill Claude CodeCodex
Fetches WhyGraph rationale — intent, constraints, tradeoffs, and risks — for existing code before it is changed, so an edit respects the original intent instead of rediscovering it.
Skill Claude CodeCodex
Show the WhyGraph rationale for a code chunk — purpose, why, constraints, tradeoffs, risks. Read-only lookup.
Skill Claude CodeCodex
Execute a reviewed WhyGraph plan markdown step by step. Reads the plan as a strict contract, runs Verify after each step, halts on first failure, appends an Implementation log to the plan.
Skill Claude CodeCodex
Plan a non-trivial code change with WhyGraph rationale + CodeGraph symbols. Asks scoping questions, then runs a fan-out/fan-in pipeline (planner → 3 parallel researchers → synthesizer) to produce a step-by-step plan.
Command
Show the WhyGraph rationale for a code chunk — purpose, why, constraints, tradeoffs, risks. Read-only lookup.
Command
Execute a reviewed WhyGraph plan markdown step by step. Reads the plan as a strict contract, runs Verify after each step, halts on first failure, appends an Implementation log to the plan.
Command
Plan a non-trivial code change with WhyGraph rationale + CodeGraph symbols. Asks scoping questions if needed, then dispatches to the planner subagent (which optionally fans out to 3 researchers + a synthesizer) to produce a step-by-step plan.
Cursor rule
When the user asks "why does X exist?", "what is X for?", "is X still needed?", "can I delete X?", or any similar intent-question about an existing code symbol, call the whygraphrationalebrief MCP tool and surface its output verbatim. Read-only — don't paraphrase or pre-answer.
Cursor rule
Always-on guidance for using the CodeGraph (codegraph) MCP tools. Prefer them over grep for structural questions; don't pull large explore output into context; trust their AST-parsed results.
Cursor rule
When the user has a reviewed WhyGraph plan markdown file and wants to execute it — "implement that plan", "execute the plan at ", "run through the steps", "carry out the plan", "let's apply the plan" — surface the /whygraph-implement slash command. Cost-gated — don't run /whygraph-implement automatically; the user…
Cursor rule
When the user wants to plan a non-trivial code change — "plan how to add X", "what would it take to migrate to Y", "outline the steps to refactor Z" — surface the /whygraph-plan slash command. The planner grounds the plan in CodeGraph (structural impact) and WhyGraph (rationale, constraints, risks). Cost-gated — don't…
Cursor rule
Before editing, refactoring, renaming, deleting, or replacing existing code, call the whygraphrationalebrief MCP tool to learn the intent, constraints, tradeoffs, and risks behind it. Skip for new code, typos, or formatter-only changes.