Run the engineering quality gate — procedure
Purpose
Use this page to run the Engineering Quality Gate: an engineering review that produces architecture classification + source-backed findings + file-specific changes.
Enforcement (fail-closed):
- Required inputs: goal, materials, constraints, and authoritative sources (or explicit permission to browse for them).
- If any required input is missing, output only:
INSUFFICIENT_EVIDENCE: <what is missing> - All non-trivial recommendations must be backed by authoritative sources; otherwise they must be removed or marked
NOT VERIFIED.
Canonical links
Prompt templates (3) Hide prompt templates
Choose a mode
- Option 1 (Standard gate run): run the gate on a small, well-scoped change set.
- Option 2 (Deep scan): run the gate on a repo snapshot / ZIP / many files with explicit coverage reporting.
- Option 3 (Claims in written text): use the technical writing gate (claims verification), not this gate.
Setup
1) Install the system prompt template:
- engineering-quality-gate.system.txt 2) Paste the runner prompt and append your inputs:
- engineering-quality-gate.user.txt 3) Provide inputs (required):
- Goal/intent
- Materials (file tree + relevant files/snippets/excerpts)
- Constraints (language/framework/runtime; non-functional constraints)
- Authoritative sources (official docs/standards/primary references) or explicit permission to browse 4) If the materials are a repo snapshot / ZIP / many files, prepend:
- deep-scan.user.txt 5) Require the output sections (exact order):
A) CONTEXT SNAPSHOTB) ARCHITECTURE CLASSIFICATION (EVIDENCE-BASED)C) FINDINGS (ACTIONABLE, SOURCE-BOUND)D) CHANGESET PLAN (FILE-SPECIFIC)E) CONFIDENCE
Verify (smoke test)
1) Run the gate with a minimal set of complete inputs (goal + 1–2 files + constraints + at least one authoritative source).
- Expected: outputs
A→Ein order. 2) Omit one required input (e.g., omit sources). - Expected: output only
INSUFFICIENT_EVIDENCE: <what is missing>
Options
Option 1 — Standard gate run
- Policy (rules): Engineering Quality Gate Policy
- System prompt template (copy/paste): engineering-quality-gate.system.txt
- User runner template (copy/paste): engineering-quality-gate.user.txt
Example
- Question: “Review this refactor for layering violations and propose minimal deltas.”
- You must provide: goal + file tree + relevant files/snippets + constraints + authoritative sources (or permission to browse).
Option 2 — Deep scan (repo snapshot / ZIP / many files)
- Policy (rules): Engineering Quality Gate Policy
- System prompt template (copy/paste): engineering-quality-gate.system.txt
- User runner template (copy/paste): engineering-quality-gate.user.txt
- Prepend (copy/paste): deep-scan.user.txt
Example
- Question: “Run an architecture + best-practice review across this repo snapshot and output a file-specific changeset plan.”
- You must provide: repo snapshot/ZIP + file tree/inventory + constraints + authoritative sources (or permission to browse). Output must include explicit coverage.
Option 3 — Claims in written text (use the other gate)
- Procedure: Evidence-Gated Technical Writing Gate — procedure
- Policy (rules): Evidence-Gated Technical Writing Policy (Claims)
Example
- Question: “Verify factual claims in this article and fail closed if citations are missing.”
- You must provide: the text + claim list (or allow extraction) + admissible citations.
Common mistakes
- Running without authoritative sources (or without explicit permission to browse), then expecting best-practice recommendations.
- Providing code snippets without file paths/structure, then expecting architecture classification.
- Mixing “claims verification for writing” into this gate (use the technical writing gate instead).
- Ignoring
INSUFFICIENT_EVIDENCEand continuing without supplying the missing inputs.