Ai persona definition

AI Persona Definition

This skill guides the creation of a complete AI Persona contractual specification using the methodology's 19-section template. The output is a filled-in definition document that serves as the formal delegation contract for an AI Persona. Supports reading persona stubs from canonical models, writing back enriched data, and batch-defining multiple personas in a domain.

When to Use

• After the Three Questions of Delegation identify a strong AI Persona candidate
• When formalising an existing AI implementation into a governed persona
• When reviewing or updating an existing persona definition
• As part of the delegation package preparation (Chapter 8)
• When batch-defining all AI Personas in a domain or company model
• When enriching persona stubs from canonical models (see also bsa-persona-enricher for automated pre-population)

Template Reference

The full template is at: .claude/agents/references/templates/AI-Persona-Definition-Template.md

Instructions

1. Prerequisites

Before starting, ensure you have:

• Completed Three Questions of Delegation results (strong answers required)
• The service architecture context (which service domain owns this persona)
• A named business owner who will be accountable
• The risk classification tier (1/2/3) from Chapter 9

1b. Canonical Model Integration (Optional)

If a canonical model exists for the persona's company, read the persona stub to pre-populate sections:

1. Load the canonical model from docs/architecture/bsa/canonical-models/[company].json
2. Find the aiPersonas entry matching the persona being defined
3. Extract: id, serviceId, authorityModel, confidenceThresholds, riskClassification, sops
4. Find the owning service for name, mission, owner, tags
5. Find all interactions where the persona's service is sourceId or targetId
6. Find all actors associated with the persona's service

This pre-populates Sections 1, 2, 4, 5 (partially), 8, 9, and 13 (from tags). Remaining sections require manual input.

2. Work Through the 19 Sections

Complete each section in order. The sections are:

#	Section	Key Inputs Needed	Auto-Fill from Model
1	AI Persona Identification	Name, ID, version, owning service	Yes
2	Governance and Ownership	Business owner, AI RM liaison, controlling domain	Partial
3	Three-Question Discovery Results	Delegation statement, management approach, team integration	No
4	Intent and Mandate	Primary intent, business objective alignment, success criteria	Partial
5	Capabilities (Permitted Actions)	Exhaustive list of CAN do; explicit list of CANNOT do	Partial (from SOPs)
6	Data Contract	Permitted sources, fields, restrictions, quality thresholds	No
7	Scope & Ways of Working	Business rules, ethical guardrails, operational constraints	No
8	Authority Model	Owner vs Invoker authority, confidence thresholds, escalation triggers	Yes
9	Interacting Services and Actors	Service dependencies, human interactions, system integrations	Yes
10	Performance Measurement (5-Tier)	Measures at each tier the persona contributes to	No
11	Multi-Contributor KPI Attribution	How shared KPIs are attributed between human and AI	No
12	Lifecycle Management	Development, deployment, monitoring, retraining schedule	No
13	Regulatory and Compliance	Applicable regulations, audit needs	Partial (from tags)
14	Risk Management	Risk assessment, mitigation controls, incident response	Partial (risk tier)
15	Trusted AI (Scope) Framework	Scope boundary definition, clarity checklist	No
16	Collaborative Data Supply Chain	Data flow mapping, upstream/downstream contracts	Partial
17	Worked Example (optional)	Scenario walkthrough	No
18	Approval and Sign-Off	Business owner, AI RM, compliance sign-off	Partial
19	Supporting Documentation	References to diagrams, specs, related personas	Yes

3. Quality Criteria for Each Section

Section 3 — Three Questions must have STRONG answers:

• Q1 (What to delegate): Specific, bounded, measurable. Weak: "Help with tasks." Strong: "Approve standard POs under $5,000 where vendor is pre-qualified and budget confirmed."
• Q2 (How to manage): Concrete oversight mechanisms. Weak: "We'll monitor it." Strong: "Dual approval for vendors near limits; alert lead when variance exceeds 10%; quarterly audit."
• Q3 (How to include): Clear operational rhythm. Weak: "It runs in background." Strong: "Reviews requests every 2 hours; lead reviews weekly; team adjusts guardrails quarterly."

Section 5 — Capabilities must be exhaustive:

• If an action is not explicitly listed as permitted, it is prohibited
• Use a capability table: ID, action, inputs, outputs, approval required

Section 8 — Authority model must be unambiguous:

• Owner Authority: Standing delegation, persona acts autonomously within mandate
• Invoker Authority: Task-specific, persona acts as assistant for human/system request
• Hybrid: Autonomous above confidence threshold, escalates below it

4. Governance Calibration by Risk Tier

Adjust governance depth based on risk classification:

Risk Tier	Governance Weight	Sections Requiring Extra Detail
Tier 1 (Low)	Light	Sections 1-8 sufficient; 12-14 can be brief
Tier 2 (Medium)	Standard	All sections complete; quarterly reviews
Tier 3 (High)	Heavy	All sections with maximum detail; monthly reviews; external audit

5. Batch Mode

When defining multiple personas across a domain or company:

1. Survey: List all AI Personas in the target scope (from canonical model aiPersonas array or from discovery outputs)
2. Prioritise: Order by risk tier (Tier 3 first) then by deployment timeline
3. Batch summary: Generate a tracking table:

# Persona Definition Batch: [Domain/Company]

| # | Persona ID | Name | Risk Tier | Status | Sections Complete | Blocker |
|---|-----------|------|-----------|--------|-------------------|---------|
| 1 | ap-quality | Quality AI | Tier 3 | In Progress | 12/19 | Awaiting Q3 answer |
| 2 | ap-demand | Demand Forecast | Tier 2 | Not Started | 0/19 | — |
| 3 | ap-monitor | Production Monitor | Tier 1 | Complete | 19/19 | — |

4. Cross-reference: As you define each persona, note interactions with other personas in the batch — these inform Section 9 and Section 16 for related personas
5. Completion: After all personas are defined, run a consistency check across the batch to ensure cross-references are reciprocal

6. Write-Back to Canonical Model

After completing a persona definition, update the canonical model with any new information:

• New SOPs identified during capability analysis (Section 5) → add to aiPersonas[].sops
• New interactions identified (Section 9) → add to interactions
• New actors identified (Section 9) → add to humanActors or systemActors
• Updated risk classification (Section 14) → update riskClassification
• Updated authority thresholds (Section 8) → update confidenceThresholds

Use canonical-model-manager incremental update patterns for the write-back.

7. Completion Checklist

Before finalising, verify:

• All 19 sections completed with specific (not generic) information
• Three Questions have strong answers
• Business owner identified and accountable
• Scope boundaries explicitly defined (in-scope AND out-of-scope)
• Capabilities list is exhaustive
• Data contract specifies exact fields, not just system names
• Authority model is clear and appropriate for risk tier
• Performance measures span all 5 tiers
• Multi-contributor attribution method defined
• Tripartite accountability confirmed (Service Owner + SOP Owner + AI RM)
• Canonical model updated with any new information (if model integration used)

Output Location

Write persona definitions alongside existing per-company persona docs: docs/architecture/bsa/[company-name]/personas/[persona-slug].md

This matches the existing structure on disk (e.g., docs/architecture/bsa/aerovista/personas/aog-response-coordinator-ai.md). Use kebab-case slugs derived from the persona name.

Batch summary (if batch mode): docs/architecture/bsa/[company-name]/personas/_batch-summary.md

Relationship to Other Skills

• Upstream: canonical-model-manager (provides persona stubs), bsa-persona-enricher (automates pre-population)
• Downstream: readiness-scorecard (requires complete definition), measurement-blueprint (Sections 10-11)
• Write-back: canonical-model-manager (update model after definition), canonical-model-validator (validate after write-back)

Output

A completed AI Persona Definition document following the 19-section template structure. This document becomes part of the delegation package alongside the SOP summary and readiness scorecard.