claw/docs/superpowers/plans/2026-04-20-generated-scene-runtime-semantics-gap-analysis-plan.md

Generated Scene Runtime Semantics Gap Analysis Plan

Status: Superseded by docs/superpowers/plans/2026-04-20-generated-scene-source-first-runtime-semantics-hardening-plan.md

Parent

• Parent design: 2026-04-20-generated-scene-runtime-semantics-gap-analysis-design.md

Goal

Analyze the 102 final generated scene skills for runtime-semantics divergence, using sweep-030-scene as the anchor case and systematizing the five gap classes exposed during inner-network validation.

This plan is analysis-only.

Fixed Inputs

• examples/scene_skill_102_final_materialization_2026-04-19/skills
• tests/fixtures/generated_scene/scene_skill_102_deterministic_invocation_readiness_after_keyword_refinement_2026-04-20.json
• tests/fixtures/generated_scene/scene_skill_102_natural_language_parameter_readiness_2026-04-20.json
• tests/fixtures/generated_scene/scene_skill_102_parameter_dictionary_template_normalization_2026-04-20.json
• Anchor source:
  • D:/desk/智能体资料/全量业务场景/一平台场景/台区线损大数据-月_周累计线损率统计分析

Boundaries

Allowed:

• Read skill manifests, reports, references, and selected source-scene evidence
• Produce JSON inventory and report

Forbidden:

• No edits in src/
• No edits to generated skills
• No rerun materialization
• No execution board updates
• No pseudo-production execution
• No implementation patch for any scene

Phase 0: Freeze Gap Taxonomy

Tasks:

1. Fix the five runtime-semantics gap classes from the anchor case
2. Define high / medium / low risk buckets
3. Lock analysis outputs and stop rule

Acceptance:

1. The five gap classes are explicit and stable
2. The plan remains analysis-only

Phase 1: Anchor-Case Evidence Extraction

Tasks:

1. Read sweep-030-scene generated assets:
   • scene.toml
   • references/generation-report.json
   • references/org-dictionary.json
   • generated script
2. Read source-scene evidence from the original 台区线损大数据-月_周累计线损率统计分析
3. Record direct evidence for:
   • alias gap
   • dictionary recovery gap
   • parameter default semantics gap
   • resolver-to-request mapping gap
   • runtime URL semantics gap

Acceptance:

1. sweep-030-scene has explicit evidence for each applicable gap class

Phase 2: 102-Scene Inventory Scan

Tasks:

1. Scan all 102 final skills
2. Extract:
   • deterministic keywords
   • params presence
   • dictionary reference presence
   • bootstrap target presence
   • generation-report URL evidence
3. Tag scenes with likely gap classes using bounded heuristics

Acceptance:

1. Every scene gets a runtime-semantics record
2. Every scene has riskLevel and gaps

Phase 3: Family / Archetype Grouping

Tasks:

1. Group findings by archetype / family
2. Count gap incidence by bucket
3. Separate:
   • generator-level fix candidates
   • runtime-only residuals

Acceptance:

1. Summary counts exist per gap type and per archetype
2. Report can distinguish generator vs runtime responsibilities

Phase 4: Publish Analysis Assets

Deliverables:

1. tests/fixtures/generated_scene/generated_scene_runtime_semantics_gap_analysis_2026-04-20.json
2. docs/superpowers/reports/2026-04-20-generated-scene-runtime-semantics-gap-analysis-report.md

Acceptance:

1. All 102 scenes are represented
2. sweep-030-scene is explicitly called out as anchor evidence
3. The report recommends next implementation routes, but does not execute them

Stop Statement

Stop after publishing the JSON inventory and report.