git:20261005.41e388b to git:20261006.f0a700f

6 added, 9 removed. Audit B to B.

---

description: "Review a game design document from an implementer's perspective — check self-consistency, playability, buildability, and codebase gaps. Use after writing a design doc, or when you want to validate an existing design against the current codebase."

---

+ > Run qq scripts as `${CLAUDE_PLUGIN_ROOT}/bin/<name>`: they are not on PATH, so a bare `qq-execute-checkpoint.py` exits 127.

Respond in the user's preferred language (detect from their recent messages, or fall back to the language setting in CLAUDE.md).

Review a game design document from an implementer's perspective.

Arguments: $ARGUMENTS (path to a design document, or empty to use the most recent design doc in `Docs/qq/`)

## Process

1. **Find the document:** if a path is given, read it. Otherwise, find the most recent `*_design.md` in `Docs/qq/`.

- 2. **Spawn a review subagent** using the Agent tool. The subagent should:

- - Read [design-reviewer-prompt.md](design-reviewer-prompt.md) for the full review checklist

- - Read the design document

- - **Independently** explore the codebase (Services, configs, existing design docs) to verify claims — do not rely solely on what the document says exists

- - Output the review in the format specified in the reviewer prompt

- 3. **Verify the findings:** as the main agent, independently spot-check the subagent's claims against the codebase. The subagent may misread code or reference stale information. For each finding, confirm or reject it with your own evidence before presenting to the user.

- 4. **Present the verified review** to the user — only include findings you confirmed. Note any subagent findings you rejected and why.

- 5. **If verdict is HAS GAPS or NEEDS REWORK:** revise the document together with the user. Re-run the subagent review after revisions. Loop until SOLID or the user explicitly accepts the gaps. Items moved to the document's "Needs the user's decision" list count as resolved.

- 6. **If verdict is SOLID:** confirm and recommend `/qq:plan`. **If invoked with `--auto` (or called from a skill running in `--auto` mode):** run `qq-execute-checkpoint.py pipeline-advance --project . --completed-skill "/qq:post-design-review" --next-skill "/qq:plan"`, then invoke `/qq:plan --auto <document-path>`.

+ 2. **Spawn a review subagent** (Agent tool) that reads [design-reviewer-prompt.md](design-reviewer-prompt.md) and the design document, checks the document's claims against the codebase itself, and returns the review in that prompt's format.

+ 3. **Verify, then present:** check each finding yourself against the document and the code; the subagent can misread code or cite stale information. Present only the findings you confirmed, and say which ones you rejected and why.

+ 4. **If verdict is HAS GAPS or NEEDS REWORK:** revise the document together with the user. Re-run the subagent review after revisions. Loop until SOLID or the user explicitly accepts the gaps. Items moved to the document's "Needs the user's decision" list count as resolved.

+ 5. **If verdict is SOLID:** confirm and recommend `/qq:plan`. **If invoked with `--auto`:** run `qq-execute-checkpoint.py pipeline-advance --project . --completed-skill "/qq:post-design-review" --next-skill "/qq:plan"`, then invoke `/qq:plan --auto <document-path>`.