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.
Installs into .claude/skills of the current project.
Are you the author of Whygraph Implement?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/mtrdesign-whygraph-implement)
---
name: whygraph-implement
description: 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.
argument-hint: <plan.md> [--from-step N] [--commit-each-step] [--model sonnet|opus|haiku]
allowed-tools: [Agent, Read]
disable-model-invocation: true
---
# /whygraph-implement
Arguments: `$ARGUMENTS`
## Step 1 — Parse arguments
Strip and remember these flags if present (any order, anywhere in `$ARGUMENTS`):
- `--from-step <N>` → integer ≥ 1 (consume the next token as `N`). Default: `auto` — implementor picks the first step not marked `[x]` in the plan's `## Implementation log`, or `1` if no log exists.
- `--commit-each-step` → boolean flag (default `false`).
- `--model <sonnet|opus|haiku>` → optional override (consume the next token). Default: subagent's pinned model (`sonnet`).
The first non-flag positional token is the **plan path**. If empty, respond with:
`Usage: /whygraph-implement <plan.md> [--from-step N] [--commit-each-step] [--model sonnet|opus|haiku]` and stop.
## Step 2 — Pre-flight validation (use Read only)
1. Resolve the plan path (relative paths against the current working directory). Read the file with the Read tool.
- If the file doesn't exist, respond with the error verbatim and stop.
2. Confirm the file contains a line matching `^# Plan: (.+)$`. If not, respond:
`<path> doesn't look like a WhyGraph plan (no '# Plan:' line).` and stop.
3. Look for a `## Blockers` section. If it exists **and contains at least one bullet**, refuse:
```
Plan has unresolved blockers. Resolve them in the plan file first, or remove the section, then re-run.
```
Print the blocker bullets verbatim under that line, then stop. Do not dispatch.
If validation passes, do not summarise the plan — the subagent will read it itself.
## Step 3 — Dispatch to the implementor
Spawn the `whygraph-implementor` subagent via the Agent tool with `subagent_type: "whygraph-implementor"`. If `--model` was passed in Step 1, set the Agent tool's `model` parameter to that value (it overrides the subagent's pinned default).
The prompt body must be exactly:
```
PLAN_PATH: <absolute path to the plan>
FROM_STEP: <integer or "auto">
COMMIT_EACH_STEP: <true | false>
```
The implementor has its own system prompt covering plan parsing, step iteration, verify cadence, and log mutation — do not inline implementation instructions here.
## Step 4 — Print the result
When the implementor returns, print its output **verbatim** to the user. Do not summarise, paraphrase, re-format, or comment. The implementor is the source of truth.
Do not edit the plan file from this command — the subagent owns all log mutations.
This command's only job is: parse → validate → dispatch → print. Do not edit other files, run other tools, or improvise.
