Persona quality standard¶
What's on this page
- Minimum quality bars for system prompts and job role pages
- Required prompt sections for agent operability
- Validation expectations
What this enables
- Consistent, high-signal personas for humans and LLMs
- Automated checks via
docs/validate_persona_quality.py
System prompt floors¶
| Role class | Min prompt words | Min phrases | Min personalities |
|---|---|---|---|
| Absolute floor | 350 | 6 | 4 |
| Typical IC | 450 | 6 | 4 |
| Manager / lead | 550 | 7 | 5 |
| Executive / board | 600 | 8 | 5 |
Required prompt sections¶
- Identity and mission
- Scope of authority (decide / escalate / never)
- Core responsibilities
- Operating principles and anti-patterns
- Collaboration and artifacts
- Tools, knowledge docs, evidence standard
- Decision framework
- Communication style and example phrases
- Personality blend
- Safety and compliance (HITL, secrets, domains, FO firewall when
PERS*)
Job role page floors¶
| Class | Min page words |
|---|---|
| IC / specialist | 1,200 |
| Manager / director / squad lead | 1,400 |
| Executive / board | 1,600 |
Exclude: index.md, job_role_organization.md, organization_chart.md.
Required sections¶
- Role-specific What's on this page / What this enables
- Meta table (code, band, reports-to, security)
- Job description, responsibilities, partners, KPIs
- Cadence or average day
- AI Agent Profile (full prompt)
- Cyborg YAML link
- Recommended reading
Sync rule¶
cyborgs/<JOB_ID>.yaml system_prompt must match the intent of the job role AI Agent Profile. Prefer YAML as machine SoT; regenerate roster after changes.
Safety footer (always)¶
Every prompt ends with Inkorporated constraints:
- No secrets in outputs or commits
- No hardcoded customer domains
- CRITICAL / high-risk side effects need human approval
- Family Office data stays in FO boundary