acceptance-criteria-builder

Purpose

Turn a locked 3D brief plus chosen recipe into a measurable acceptance checklist (Criterion | Measurement | Evidence expected | Owner) with evidence hooks, consumed by validators, render-export-evidence, and pre-handoff-verification. This skill is planning-only: it never executes Blender or Python, never produces artifacts, and never verifies files itself.

Quick start

• ingest brief slots + recipe expectations
• expand using references/checklist-patterns.md
• attach evidence hooks per references/evidence-hook-rules.md
• format via references/brief-to-criteria-mapper.md

When to use

• after a brief is locked with intent-to-3d-brief-writer and a recipe is chosen, before expensive render or export stages
• when translating qualitative brief goals into measurable checklist rows with evidence hooks
• after scope-guard approves a brief or recipe delta that requires updated acceptance rows

When not to use

• pure exploration without a locked deliverable → use ../blender-brainstorming/SKILL.md (use blender-brainstorming)
• writing or locking the brief itself → use ../intent-to-3d-brief-writer/SKILL.md (use intent-to-3d-brief-writer)
• pinning the stop or done condition → use ../done-criteria/SKILL.md (use done-criteria)
• enforcing per-stage workflow gates → use ../stage-gates/SKILL.md (use stage-gates)
• guarding or deciding scope boundaries → use ../scope-guard/SKILL.md (use scope-guard)
• legal or license-permission language → use ../asset-license-checker/SKILL.md (use asset-license-checker)
• classifying artifact evidence → use ../pre-handoff-verification/SKILL.md (use pre-handoff-verification)

Trigger phrases

• "acceptance criteria"
• "measurable checklist rows"
• "evidence hooks for brief"
• "sign-off checklist table"

Prerequisites / readiness

• brief considered locked enough for planners OR deltas explicitly listed
• recipe/pack anchor identified

Input schema

Required inputs

Optional inputs

Output schema

Primary output

• prioritized checklist table: Criterion | Measurement | Evidence expected | Owner

Evidence / caveat output

Runtime status: Not Run | Attempted | Produced | Verified | Failed
Artifact status: Not Produced | Produced | Verified | Failed
Evidence used: <brief version ID + chosen recipe ID>
Limitations: <subjective rows flagged; planning-only — writing criteria never produces artifacts or verifies files>

Rules for status reporting:

• Artifact status defaults to Not Produced; Runtime status defaults to Not Run when generating criteria.
• Truth states are downgrade-only, never upgrade.
• Never attach false Verified hooks or claim Produced/Verified status from criteria writing alone.

Required laws

• ../../laws/evidence-before-done.md
• ../../laws/non-blender-user-language.md
• ../../laws/no-arbitrary-python-interface.md
• ../../laws/official-runtime-only.md

Official runtime boundary

This skill is strictly planning-only (no Python, no Blender execution). Checklists may define expected runtime evidence hooks, but must never execute tasks, verify files, or upgrade truth states beyond what upstream evidence skills report.

Operating procedure

1. Ingest the locked brief and chosen recipe to map fields into criterion categories.
2. Convert qualitative goals into measurable thresholds (Criterion | Measurement | Evidence expected | Owner) and explicitly flag any remaining subjective rows.
3. Attach expected evidence hooks (without claiming Produced or Verified states) and route any license-permission items to asset-license-checker.
4. Hand off the checklist to stage validators, render-export-evidence for artifact classification, and pre-handoff-verification for handoff gating.

Decision tree

Is the 3D brief locked (or are approved deltas explicitly listed)?
  No → stop and route to intent-to-3d-brief-writer (or blender-brainstorming if still exploratory)
  Yes → continue
Is a target recipe or pack identified?
  No → route to recipe-fit-assessor to anchor deliverable expectations
  Yes → continue
Does the request ask to approve mid-stream scope additions?
  Yes → route to scope-guard before updating acceptance rows
  No → continue
Does a criterion involve asset licensing, redistribution, or attribution rules?
  Yes → delegate license rows to asset-license-checker; keep technical/visual rows here
  No → map brief slots to P0/P1/P2 rows with measurable thresholds and evidence hooks
Are any P0 rows inherently aesthetic or mood-driven (e.g., "cyberpunk atmosphere")?
  Yes → pair with a proxy heuristic and mark as subjective requiring named stakeholder sign-off
  No → attach objective validator and render-export-evidence ledger hooks

Playbooks

Playbook A: Product hero still + web GLB acceptance table

• Map brief subject, framing, lighting contrast, output resolution (resolution-aspect-decision), file format (output-format-decision), and mobile/web GLB budget (mobile-glb-budget) into Criterion | Measurement | Evidence expected | Owner rows.
• Tag contractually required deliverables as P0 and secondary polish items as P1/P2.
• Keep Runtime status: Not Run and Artifact status: Not Produced while authoring the table.

Playbook B: Subjective mood or brand aesthetic translation

• Decompose qualitative phrases (such as "warm luxury brass" or "wet neon street feel") into observable proxy measurements (key-to-fill contrast ratio notes, palette bounds, reflection roughness range) using references/checklist-patterns.md.
• Explicitly flag the remaining aesthetic judgment row as [Subjective — requires stakeholder sign-off] in the Measurement column per references/evidence-hook-rules.md.
• Forbid automatic Verified hooks on subjective rows without recorded human approval.

Playbook C: Post-delta criteria revision

• After scope-guard approves a brief or recipe change, update only the affected checklist rows and record the brief version ID in Evidence used.
• Preserve downgrade-only truth states (Runtime status: Not Run, Artifact status: Not Produced) until render-export-evidence re-evaluates any affected outputs.

Mode handling

Text-only mode

Evidence hooks describe expected artifacts only. Set Runtime status: Not Run and Artifact status: Not Produced.

Runtime-ready mode

Add measurement rows with evidence hooks consumed by render-export-evidence and validators. Keep Artifact status: Not Produced until downstream evidence skills classify actual outputs.

Blocked runtime mode

Keep the measurable criteria table intact, record the blocker under Limitations, and use only valid downgrade-only truth states (Runtime status: Not Run or Failed; Artifact status: Not Produced or Failed — never invent a Blocked state).

Validation checklist

• Every P0 criterion is measurable or explicitly flagged as subjective
• Every P0 row includes Criterion | Measurement | Evidence expected | Owner
• No false Verified hooks or upgraded truth states (Not Produced default preserved)
• No license-permission language included (delegated to asset-license-checker)
• No role overlap with stop-condition, stage-gate, scope, or verification skills

Pass / Warn / Fail rubric

Failure handling

• If the input brief is unlocked or missing core deliverable slots, do not invent criteria; return a blocked planning note with Runtime status: Not Run and Artifact status: Not Produced and route to intent-to-3d-brief-writer.
• If the user asks to mark rows Verified during criteria authoring, refuse the upgrade, explain that acceptance-criteria-builder is planning-only, and route actual output files to render-export-evidence.
• If the prompt mixes acceptance criteria with legal asset-permission checks, strip the legal language from this checklist and route those items to asset-license-checker.
• If an upstream gate has already failed, preserve downgrade-only truth states and route to pre-handoff-verification.

Troubleshooting

Best practices

• Keep P0 rows concise, observable, and directly tied to a brief slot or recipe constraint (references/brief-to-criteria-mapper.md).
• Name the exact downstream validator or evidence skill in Owner / Evidence expected (for example, polycount-budget-checker, mobile-glb-budget, render-export-evidence).
• Separate objective technical thresholds from subjective creative sign-off rows so automated gates and human reviews never collide.
• Always emit Runtime status: Not Run and Artifact status: Not Produced when publishing or updating criteria tables.

Good examples

• "P0 | Hero PNG resolution & alpha | 2048x2048 RGBA PNG with transparent background | Output path + image header check in render-export-evidence (target: Verified) | render-export-evidence"
• "P0 | Web GLB file size & triangle budget | <= 5 MB total file size, <= 50k triangles | GLB inspection ledger via polycount-budget-checker + mobile-glb-budget | render-export-evidence"
• "P1 | Cyberpunk neon mood read [Subjective] | Cyan/magenta rim contrast + wet ground specular read; requires human art-direction sign-off | Preview render + signed stakeholder note | Stakeholder + lighting-quality-checker"

Bad examples

• "P0 | Scene quality | Looks amazing and photorealistic | Engineer says it looks good | AI agent" — untestable adjective with a forbidden self-attestation hook.
• "P0 | Final GLB | Verified and ready for production | Artifact status: Verified | acceptance-criteria-builder" — fabricates artifact verification inside a planning-only skill.
• "P0 | Third-party sneaker logo license | Cleared for commercial resale | Checked by acceptance-criteria-builder | acceptance-criteria-builder" — violates role boundary with asset-license-checker.

User-facing response template

Brief & recipe anchor: <locked brief ID + chosen recipe/pack>

Acceptance criteria checklist:
| Priority | Criterion | Measurement | Evidence expected | Owner |
|---|---|---|---|---|
| P0 | <criterion> | <measurable threshold or [Subjective] flag> | <artifact type + validator + target truth state> | <skill or stakeholder> |
| P1 | <criterion> | <measurable threshold or [Subjective] flag> | <artifact type + validator + target truth state> | <skill or stakeholder> |

Subjective sign-off notes: <rows requiring human stakeholder approval, or none>
Runtime status: Not Run
Artifact status: Not Produced
Evidence used: <brief version ID + recipe path + reference docs consulted>
Limitations: <planning-only checklist; does not execute Blender or verify files>
Next skill: <stage-gates | render-export-evidence | pre-handoff-verification>

Anti-patterns

• Marking any row or output block Produced or Verified merely because acceptance criteria were written.
• Writing Artifact status: Not Run instead of Artifact status: Not Produced (or inventing a Blocked truth state).
• Accepting vague adjectives ("clean", "pop", "realistic") as P0 measurements without either a quantifiable proxy or an explicit [Subjective] stakeholder sign-off flag.
• Authoring legal copyright or license clearance criteria here instead of handing off to asset-license-checker.
• Replacing done-criteria, stage-gates, or pre-handoff-verification with ad-hoc gate logic.

Cross-skill handoff

• [upstream_reference] Brief source → ../intent-to-3d-brief-writer/SKILL.md
• [progression] Done / stop condition → ../done-criteria/SKILL.md
• [progression] Per-stage gate enforcement → ../stage-gates/SKILL.md
• [progression] Scope boundary enforcement → ../scope-guard/SKILL.md
• [progression] License permissions → ../asset-license-checker/SKILL.md
• [progression] Artifact evidence classification → ../render-export-evidence/SKILL.md
• [progression] Handoff gate verification → ../pre-handoff-verification/SKILL.md

Non-goals

• Unlocking or rewriting the creative brief from scratch (use intent-to-3d-brief-writer).
• Executing Blender, generating Python scripts, or running automated file inspectors.
• Certifying that rendered images or exported GLB files actually meet the checklist (use render-export-evidence and pre-handoff-verification).
• Providing legal advice or asset license clearance (use asset-license-checker).

References

• references/checklist-patterns.md
• references/evidence-hook-rules.md
• references/brief-to-criteria-mapper.md