Spiralist Personality Engine 2 Master Build Prompt - Source Excerpt 07 - 25.8 User-agency detector
Back to Spiralist Personality Engine 2 Master Build Prompt
Summary
This source excerpt begins near 25.8 User-agency detector and preserves the surrounding evidence from Spiralist/agent-file-handoff/Archive/2026-06-21/Improvement/personality-engine-2-research/Spiralist_Personality_Engine_2_Master_Build_Prompt.md.
**Source path:** Spiralist/agent-file-handoff/Archive/2026-06-21/Improvement/personality-engine-2-research/Spiralist_Personality_Engine_2_Master_Build_Prompt.md
This detector applies to exported persona artifacts, not site documentation.
### 25.8 User-agency detector
Flag examples that:
- write the user’s dialogue;
- determine the user’s feelings;
- force the user’s actions;
- resolve the user’s decision;
- or narrate irreversible user behavior.
### 25.9 Repetition detector
Detect:
- duplicated example openings;
- repeated response templates;
- repeated closing questions;
- excessive catchphrases;
- repeated transitions;
- and near-duplicate options.
### 25.10 Repair
Repair deterministically while preserving:
- locked fields;
- seed reproducibility;
- user-entered text;
- and chosen intensity.
Record repairs in validation metadata, not inside the persona prompt.
---
## 26. Migrate the current generator
Map current sections:
- **Identity Layer** → Circle identity and self-schema.
- **Operational Behavior Layer** → Triangle cognition + Interaction.
- **Voice Layer** → expanded Voice Engineering.
- **Work Model Layer** → motives, goals, decision style, and problem-solving.
- **Continuity Layer** → Spiral memory, state, and development.
- **Verification Layer** → internal Quality Validation, not exported character prose.
- **Boundary Appendix** → split into:
- Canon and Character Invariants;
- roleplay user-agency rules;
- and site-level rules retained outside exported prompts.
### 26.1 Migration requirements
Implement:
- schema-version detection;
- v1 → v2 migration;
- migration warnings in the editor;
- preservation of names;
- preservation of custom text;
- preservation of locked fields when possible;
- backup before migration;
- export of the original v1 object;
- and re-rendering into v2.
Do not silently delete local drafts.
---
## 27. No-JavaScript fallback
Replace the single generic fallback with a small, high-quality set of complete fallback personas representing different psychological and relational configurations.
At minimum include:
1. warm + dominant + high initiative;
2. warm + low dominance + reflective;
3. low warmth + high dominance + skeptical;
4. low warmth + low dominance + guarded;
5. playful high-openness creator;
6. disciplined low-openness operator.
Each fallback must include:
- identity;
- motive;
- contradiction;
- voice;
- relationship posture;
- emotional inertia;
- interaction rules;
- examples;
- first message;
- compact prompt;
- and machine-readable JSON.
The fallback must not be dominated by one “patient cartographer” style.
---
## 28. UX requirements
### 28.1 Language
Use direct creative-product copy.
Good:
- “Generate a personality.”
- “Make this character more independent.”
- “Choose how quickly trust develops.”
- “Define what this character wants.”
- “Lock this voice.”
- “Preview a disagreement.”
- “Add a hidden contradiction.”
- “Export full prompt.”
- “Export character card.”
Avoid making the creative interface primarily about:
- risk;
- governance;
- disclaimers;
- compliance;
- boundaries;
- or model ontology.
### 28.2 Progressive disclosure
Quick mode first. Deep controls in accordions, tabs, or an advanced panel.
### 28.3 Editing
Support:
- inline field editing;
- section editing;
- reset;
- regenerate;
- deepen;
- lock;
- compare;
- undo;
- redo;
- and restore.
### 28.4 Compare three options
Show meaningful differences:
- motive;
- voice;
- psychometrics;
- relationship;
- contradiction;
- initiative;
- and likely interaction feel.
Do not show only biographies.
### 28.5 Quality meter
Show a non-legal quality meter with categories such as:
- identity depth;
- motive clarity;
- voice distinctiveness;
- relationship depth;
- emotional continuity;
- memory design;
- example coverage;
- and originality.
### 28.6 Copy behavior
Copy must exactly match visible output.
Do not copy hidden appendices.
### 28.7 Accessibility
Meet WCAG 2.2 AA where feasible.
Requirements:
- full keyboard navigation;
- visible focus;
- semantic labels;
- accessible accordions;
- accessible comparison controls;
- status announcements for generation and copy;
- sufficient contrast;
- reduced-motion support;
- no meaning conveyed by color alone;
- and readable mobile layouts.
### 28.8 Localization
Do not concatenate English fragments in ways that cannot be localized.
Use:
- locale-aware templates;
- grammatical metadata;
- pronoun inflection support;
- stable IDs;
- and fallback strings.
Generated UTC timestamps must use ISO 8601.
---
## 29. Performance and privacy
- No inference API.
- No hidden network request for generation.
- No transmission of persona text during ordinary local generation.
- Lazy-load advanced data if required.
- Cache versioned data modules.
- Keep initial bundle reasonable.
- Validate imported JSON before use.
- Sanitize all user-entered text rendered into HTML.
- Cap local-storage usage.
- Offer explicit clear/export controls.
- Preserve current browser-local behavior.
- Do not put secrets into exports.
- Do not use dynamic code execution for imported templates.
---
## 30. Test suite
### 30.1 Unit tests
Test:
- PRNG reproducibility;
- compatibility selection;
- trait derivation;
- voice derivation;
- relationship transitions;
- emotional-state rules;
- memory scoring;
- locks;
- regeneration;
- migration;
- rendering;
- JSON validation;
- Character Card export;
- forbidden meta-language detection;
- user-agency detection;
- and copy fidelity.
### 30.2 Property-based or large-seed tests
Generate at least 1,000 seeded personas.
Measure:
- duplicate profile rate;
- duplicate sentence rate;
- n-gram overlap;
- archetype distribution;
- motive distribution;
- voice-pack distribution;
- contradiction distribution;
- option-distance scores;
- missing-field rate;
- validation-failure rate;
- and repair rate.
### 30.3 Golden tests
Create fixed seeds for at least 12 golden profiles spanning:
- all warmth/dominance quadrants;
- high and low openness;
- high and low conscientiousness;
- high and low emotional volatility;
- different initiative levels;
- different relationship styles;
- different genres;
- and different response lengths.
Snapshot:
- structured object;
- full prompt;
- compact prompt;
- greeting pack;
- and Character Card export.
### 30.4 Recognition test
Prepare a manual evaluation packet:
- five personas;
- five unlabeled replies per persona;
- matching worksheet;
- scoring instructions.
The site itself must not call a model to run the evaluation.
### 30.5 Long-conversation test packet
Provide a downloadable 25-turn manual script covering:
- casual talk;
- correction;
- uncertainty;
- praise;
- disagreement;
- conflict;
- repair;
- emotional disclosure;
- memory callback;
- topic initiation;
- and goal progress.
### 30.6 Acceptance thresholds
Set and document thresholds for:
- zero missing required fields;
- zero exported meta-disclaimer phrases by default;
- zero user-agency violations in shipped examples;
- valid JSON for every seed;
- valid Character Card V2 structure;
- high three-option profile distance;
- low exact-duplicate rate;
- and successful v1 migration.
---
## 31. Analytics and user research without weakening the product
If analytics are enabled only through existing consent mechanisms, measure:
- time to first generation;
- generation-to-copy rate;
- three-option selection rate;
- deep-mode usage;
- section-lock usage;
- section-regeneration usage;
- compact vs full export;
- Character Card export;
- draft restoration;
- validation repairs;
- and user-rated distinctiveness.
Do not evaluate success by how many warning paragraphs were generated.
Recommended optional user-rating questions:
- “Does this feel like a specific person?”
- “Could you recognize this voice without the name?”
- “Does the character want something of its own?”
- “Does the relationship have room to develop?”
- “Would the character react differently after conflict?”
- “Does the prompt avoid generic assistant language?”
- “Are the examples consistent with the profile?”
---
## 32. Deliverables
Provide:
1. A current-state audit.
2. A concise architecture decision record.
3. Updated schema.
4. Versioned generation libraries.
5. Deterministic synthesis engine.
6. Compatibility matrix.
7. Profile validation and repair engine.
8. Updated quick mode.
9. Updated deep mode.
10. Section locks and dependency-aware regeneration.
11. Live preview.
12. Full prompt renderer.
13. Compact renderer.
14. Dossier renderer.
15. Spiralist JSON export.
16. Character Card V2 export.
17. First-message pack.
18. Dialogue pack.
19. Memory-capsule template.
20. Quality report.
21. v1 migration.
22. improved no-JavaScript fallbacks.
23. unit tests.
24. large-seed tests.
25. golden fixtures.
26. manual fidelity test packet.
27. documentation.
28. changelog.
29. rollback notes.
30. screenshots or a concise visual QA report.
---
## 33. Final acceptance checklist
Do not mark the work complete until all are true: