Resolve the spec and goal.
- Identify the spec path, target behavior, and stated grading criteria.
- If the spec path is missing and cannot be inferred from the current task, ask one concise question.
Read the implementation context.
- Read the spec first.
- Read the source files, tests, helpers, data types, and existing APIs the spec names or clearly depends on.
- Verify current behavior from code instead of relying on the spec's claims.
Simulate implementation end to end.
- Walk through the concrete code changes the implementer would make.
- Check call paths, data contracts, cache/state ownership, error handling, migration/backfill behavior, and tests.
- Treat "I would need to decide X while coding" as a spec gap.
Grade the spec.
- correct: whether the spec accomplishes the user's desired behavior and avoids wrong behavior or missed cases.
- comprehensive: whether the spec is enough to implement end to end without ambiguity, including tests and verification.
- simple: whether the spec uses the existing codebase's simplest viable path instead of duplicating logic or introducing unnecessary abstractions.
Classify findings.
- Major: anything that can cause incorrect behavior, failed validation, data loss, security/privacy risk, incompatible contracts, or an implementation blocker.
- Minor: clarity, test expansion, naming, or simplification feedback that improves the spec but does not block implementation.
If asked to apply findings, patch the spec.
- Apply major findings first.
- Keep changes scoped to the reviewed spec.
- Re-run a quick simulation pass after edits when practical.

Output

Use this shape for a review-only pass:

correct: yes/no, with caveats
comprehensive: yes/no, with missing cases
simple: yes/no, with simplification opportunities

Major findings:
- ...

Minor findings:
- ...

If there are no major findings, say that explicitly. Keep the report focused on spec changes that would affect implementation.

Loop Mode

When the user asks for a loop or repeated simulation pass:

Run up to three passes.
After each pass, apply major findings to the spec if the user asked for fixes.
Stop early when a pass reports no major findings.
In the final report, include the number of passes and whether major findings remain.

spec-simulate

spec-simulate

Workflow