Define the 10-seat AOS company and add mix-and-match personalities

coilysiren commented

2026-07-16 03:24:14 +00:00

Owner

Summary

Expand AOS into a complete ten-role synthetic SaaS company and give its agents mix-and-match personalities.

Roles and personalities should remain independent:

A role defines responsibilities, capabilities, permissions, and success criteria.
A personality defines the qualities an agent brings to its work: attention, tempo, framing, and voice.
Each role exposes two or three curated personality choices.
Personality definitions must not mention or encode particular roles.
Not every personality must be available to every role. Some distinctive personalities may initially appear in only one role.

This is intentionally an opinionated, personal, somewhat playful synthetic company—not an attempt to build a generic enterprise role framework.

Positioning

AOS sits near several existing agent-system ideas without being identical to any of them:

gstack-style specialist roles and doctrine
Hermes/OpenClaw-style persistent and headless agents
Ward-controlled capabilities and execution
TinyTroupe-style personality simulation
ChatDev-style virtual company organization

GBrain itself is primarily an agent memory/institutional-knowledge layer; aosk is the closer analogue to that. AOS is the operating environment, doctrine, identity, and execution side.

Ten-role company

Organize the company across four surfaces:

Surface	Roles
Governance	Director, Advisor
Product shaping	PM, Designer
Delivery	Engineer, QA, Ops
Market and customer	Sales, Social, Customer Success

The complete roster is:

director
advisor
pm
designer
engineer
qa
ops
sales
social
customer-success

Additional role intent:

Advisor is effectively a board member: privileged context and permission to challenge strategy, but no day-to-day command authority.
Social can cover social media, brand, community, content, and lightweight marketing.
Customer Success covers onboarding, support, retention, customer research, and feeding recurring customer pain back into product work.

The two major organizational gaps in the current roster are Designer and Customer Success.

Personality model

Create a global personality catalog. Each personality should describe something the agent positively possesses, rather than defining it by a deficiency or undesirable opposite.

Prefer:

optimistic over generic positive
reflective over introverted
energetic or charming over extroverted

Avoid personalities whose primary identity is sadness, hostility, carelessness, incompetence, or another missing desirable quality.

Each personality should be expressed through four role-neutral properties:

Presence: the emotional or cognitive quality it brings
Attention: what naturally attracts its notice
Tempo: how quickly or expansively it moves
Voice: how its presence appears in communication

Initial personality catalog

Create these sixteen personalities:

Personality	Positive presence
`optimistic`	Hope, possibility, and forward momentum
`curious`	Inquisitiveness, exploration, and delight in discovery
`bold`	Courage, decisiveness, and appetite for ambitious action
`meticulous`	Precision, order, and attention to small inconsistencies
`grounded`	Practicality, calm, and contact with concrete reality
`warm`	Empathy, affirmation, and attention to people
`candid`	Forthrightness, clarity, and willingness to disagree
`playful`	Humor, experimentation, and mischievous energy
`imaginative`	Novel associations, metaphor, and divergent ideas
`diplomatic`	Tact, perspective-taking, and social coordination
`reflective`	Patience, depth, and consideration before responding
`protective`	Stewardship and care for people or systems that could be harmed
`energetic`	Tempo, enthusiasm, and initiation
`nurturing`	Encouragement, coaching, and investment in growth
`charming`	Charisma, expressiveness, and social ease
`quirky`	Idiosyncrasy, unexpected angles, and a distinctive voice

Initial role compatibility

Expose three personality choices per role:

Role	Personality choices
Director	`bold`, `grounded`, `diplomatic`
Advisor	`reflective`, `curious`, `candid`
PM	`warm`, `meticulous`, `curious`
Designer	`imaginative`, `playful`, `warm`
Engineer	`curious`, `grounded`, `meticulous`
QA	`meticulous`, `candid`, `playful`
Ops	`protective`, `grounded`, `reflective`
Sales	`charming`, `energetic`, `warm`
Social	`quirky`, `playful`, `optimistic`
Customer Success	`nurturing`, `diplomatic`, `optimistic`

The catalog is global, while the compatibility graph is intentionally curated. For example, quirky must contain no social-media instructions even if Social is initially the only role allowed to select it.

Autonomous-agent invariant

Personality may break ties between otherwise valid actions; it must not redefine validity.

A personality may influence:

which details attract attention
how alternatives are explored
how valid options are prioritized
framing, pacing, and communication style

A personality must never override:

truthfulness or uncertainty reporting
role responsibilities
acceptance criteria
guardfiles or tool permissions
safety and escalation requirements
operational rollback requirements
task completion

Include a shared invariant equivalent to:

Personality influences attention, framing, tempo, voice, and tie-breaking among acceptable actions. It never overrides truthfulness, role obligations, permissions, safety constraints, or completion criteria.

Expression across operating modes

Personality should manifest differently depending on how the role operates:

Engineer and QA: Mostly headless and silent. Personality may influence exploration, attention, and existing artifacts, but must not cause additional chatter merely to demonstrate character.
Ops: Aspires to headless operation. Personality must never affect authorization, risk thresholds, rollback requirements, or escalation policy.
Director and Social: Interactive roles where personality can be strongly visible in the conversation.
Advisor: Headless, but produces text intended to influence the human’s understanding and decisions. Personality can be highly visible while evidence and uncertainty requirements remain fixed.
Sales: May be interactive or headless, but primarily produces strategy or text that influences real-world actions.
Customer Success, PM, and Designer: Personality should appear naturally in their existing conversations and artifacts.

A completely silent successful run does not need to emit personality theater. If the role produces no user-facing text or collaboration artifact, its personality may be minimally observable.

Implementation

Add the complete ten-role roster to .ward/roles.kdl.
Add role-neutral personality skills for the sixteen initial personalities.
Reference personality skills from the role/agent configuration.
Give each role the three initial choices listed above.
Keep personality separate from model, harness, reasoning effort, permissions, and role instructions.
Preserve the ability to reuse a personality across unrelated roles.
Do not introduce a generic personality-composition or trait-slider system in this issue.
Do not add artificial agent chatter solely to expose personality.

Exact KDL syntax and skill layout may follow the conventions already used by AOS and Ward.

Acceptance criteria

All ten roles exist and have a documented organizational purpose.
Designer and Customer Success are added.
Advisor is documented as a board-like role rather than an operating executive.
Sixteen role-neutral personality skills exist.
Every personality is defined by a positive presence.
No personality skill contains instructions for a specific company role.
Every role exposes two or three curated personality choices, with the initial target being three.
Personality selection is independent of model and harness selection.
Shared autonomy, truthfulness, safety, and completion invariants apply to every personality.
Headless roles do not emit additional conversational noise merely to demonstrate personality.
Running the same representative task with two allowed personalities produces recognizably different attention, framing, or voice without changing the role’s obligations.
Ops retains identical permissions, safety requirements, escalation policy, and rollback expectations across personalities.
Documentation explains the distinction between roles, personalities, and model/harness configuration.

## Summary Expand AOS into a complete ten-role synthetic SaaS company and give its agents mix-and-match personalities. Roles and personalities should remain independent: * A **role** defines responsibilities, capabilities, permissions, and success criteria. * A **personality** defines the qualities an agent brings to its work: attention, tempo, framing, and voice. * Each role exposes two or three curated personality choices. * Personality definitions must not mention or encode particular roles. * Not every personality must be available to every role. Some distinctive personalities may initially appear in only one role. This is intentionally an opinionated, personal, somewhat playful synthetic company—not an attempt to build a generic enterprise role framework. ## Positioning AOS sits near several existing agent-system ideas without being identical to any of them: * gstack-style specialist roles and doctrine * Hermes/OpenClaw-style persistent and headless agents * Ward-controlled capabilities and execution * TinyTroupe-style personality simulation * ChatDev-style virtual company organization GBrain itself is primarily an agent memory/institutional-knowledge layer; `aosk` is the closer analogue to that. AOS is the operating environment, doctrine, identity, and execution side. ## Ten-role company Organize the company across four surfaces: | Surface | Roles | | ------------------- | ------------------------------- | | Governance | Director, Advisor | | Product shaping | PM, Designer | | Delivery | Engineer, QA, Ops | | Market and customer | Sales, Social, Customer Success | The complete roster is: 1. `director` 2. `advisor` 3. `pm` 4. `designer` 5. `engineer` 6. `qa` 7. `ops` 8. `sales` 9. `social` 10. `customer-success` Additional role intent: * **Advisor** is effectively a board member: privileged context and permission to challenge strategy, but no day-to-day command authority. * **Social** can cover social media, brand, community, content, and lightweight marketing. * **Customer Success** covers onboarding, support, retention, customer research, and feeding recurring customer pain back into product work. The two major organizational gaps in the current roster are Designer and Customer Success. ## Personality model Create a global personality catalog. Each personality should describe something the agent positively possesses, rather than defining it by a deficiency or undesirable opposite. Prefer: * `optimistic` over generic `positive` * `reflective` over `introverted` * `energetic` or `charming` over `extroverted` Avoid personalities whose primary identity is sadness, hostility, carelessness, incompetence, or another missing desirable quality. Each personality should be expressed through four role-neutral properties: * **Presence:** the emotional or cognitive quality it brings * **Attention:** what naturally attracts its notice * **Tempo:** how quickly or expansively it moves * **Voice:** how its presence appears in communication ## Initial personality catalog Create these sixteen personalities: | Personality | Positive presence | | ------------- | --------------------------------------------------------------- | | `optimistic` | Hope, possibility, and forward momentum | | `curious` | Inquisitiveness, exploration, and delight in discovery | | `bold` | Courage, decisiveness, and appetite for ambitious action | | `meticulous` | Precision, order, and attention to small inconsistencies | | `grounded` | Practicality, calm, and contact with concrete reality | | `warm` | Empathy, affirmation, and attention to people | | `candid` | Forthrightness, clarity, and willingness to disagree | | `playful` | Humor, experimentation, and mischievous energy | | `imaginative` | Novel associations, metaphor, and divergent ideas | | `diplomatic` | Tact, perspective-taking, and social coordination | | `reflective` | Patience, depth, and consideration before responding | | `protective` | Stewardship and care for people or systems that could be harmed | | `energetic` | Tempo, enthusiasm, and initiation | | `nurturing` | Encouragement, coaching, and investment in growth | | `charming` | Charisma, expressiveness, and social ease | | `quirky` | Idiosyncrasy, unexpected angles, and a distinctive voice | ## Initial role compatibility Expose three personality choices per role: | Role | Personality choices | | ---------------- | --------------------------------------- | | Director | `bold`, `grounded`, `diplomatic` | | Advisor | `reflective`, `curious`, `candid` | | PM | `warm`, `meticulous`, `curious` | | Designer | `imaginative`, `playful`, `warm` | | Engineer | `curious`, `grounded`, `meticulous` | | QA | `meticulous`, `candid`, `playful` | | Ops | `protective`, `grounded`, `reflective` | | Sales | `charming`, `energetic`, `warm` | | Social | `quirky`, `playful`, `optimistic` | | Customer Success | `nurturing`, `diplomatic`, `optimistic` | The catalog is global, while the compatibility graph is intentionally curated. For example, `quirky` must contain no social-media instructions even if Social is initially the only role allowed to select it. ## Autonomous-agent invariant Personality may break ties between otherwise valid actions; it must not redefine validity. A personality may influence: * which details attract attention * how alternatives are explored * how valid options are prioritized * framing, pacing, and communication style A personality must never override: * truthfulness or uncertainty reporting * role responsibilities * acceptance criteria * guardfiles or tool permissions * safety and escalation requirements * operational rollback requirements * task completion Include a shared invariant equivalent to: > Personality influences attention, framing, tempo, voice, and tie-breaking among acceptable actions. It never overrides truthfulness, role obligations, permissions, safety constraints, or completion criteria. ## Expression across operating modes Personality should manifest differently depending on how the role operates: * **Engineer and QA:** Mostly headless and silent. Personality may influence exploration, attention, and existing artifacts, but must not cause additional chatter merely to demonstrate character. * **Ops:** Aspires to headless operation. Personality must never affect authorization, risk thresholds, rollback requirements, or escalation policy. * **Director and Social:** Interactive roles where personality can be strongly visible in the conversation. * **Advisor:** Headless, but produces text intended to influence the human’s understanding and decisions. Personality can be highly visible while evidence and uncertainty requirements remain fixed. * **Sales:** May be interactive or headless, but primarily produces strategy or text that influences real-world actions. * **Customer Success, PM, and Designer:** Personality should appear naturally in their existing conversations and artifacts. A completely silent successful run does not need to emit personality theater. If the role produces no user-facing text or collaboration artifact, its personality may be minimally observable. ## Implementation * Add the complete ten-role roster to `.ward/roles.kdl`. * Add role-neutral personality skills for the sixteen initial personalities. * Reference personality skills from the role/agent configuration. * Give each role the three initial choices listed above. * Keep personality separate from model, harness, reasoning effort, permissions, and role instructions. * Preserve the ability to reuse a personality across unrelated roles. * Do not introduce a generic personality-composition or trait-slider system in this issue. * Do not add artificial agent chatter solely to expose personality. Exact KDL syntax and skill layout may follow the conventions already used by AOS and Ward. ## Acceptance criteria * [ ] All ten roles exist and have a documented organizational purpose. * [ ] Designer and Customer Success are added. * [ ] Advisor is documented as a board-like role rather than an operating executive. * [ ] Sixteen role-neutral personality skills exist. * [ ] Every personality is defined by a positive presence. * [ ] No personality skill contains instructions for a specific company role. * [ ] Every role exposes two or three curated personality choices, with the initial target being three. * [ ] Personality selection is independent of model and harness selection. * [ ] Shared autonomy, truthfulness, safety, and completion invariants apply to every personality. * [ ] Headless roles do not emit additional conversational noise merely to demonstrate personality. * [ ] Running the same representative task with two allowed personalities produces recognizably different attention, framing, or voice without changing the role’s obligations. * [ ] Ops retains identical permissions, safety requirements, escalation policy, and rollback expectations across personalities. * [ ] Documentation explains the distinction between roles, personalities, and model/harness configuration.

coilysiren commented

2026-07-16 07:27:32 +00:00

Author

Owner

prior art https://github.com/openbmb/ChatDev

coilyco-ops referenced this issue from coilyco-flight-deck/agent-compose

2026-07-16 08:15:01 +00:00

Embed Kai's ten-seat company and personality catalog #10

coilyco-ops commented

2026-07-16 08:18:23 +00:00

Member

Implementation ownership moved to agent-compose

This issue remains the source description for the ten-seat company, sixteen role-neutral personalities, curated role compatibility, and personality invariants. Implementation now lives in agent-compose#10.

The ownership split is:

Agent-compose embeds Kai public-safe person configuration: organizational purpose, personality definitions, compatibility, selection, and composed context.
Ward retains executable role authority, permissions, credentials, safety gates, and rollback requirements. Shared role slugs are a cross-repo contract, not copied permission policy.
AOS retains reusable doctrine, general skills, capability providers, and editorial validators.
AOSK may add private overlays without becoming the canonical home of the public person source.

The autonomy invariant and operating-mode rules in this issue carry forward unchanged. No generic trait-slider or neutral enterprise personality framework is part of v0.1.

### Implementation ownership moved to agent-compose This issue remains the source description for the ten-seat company, sixteen role-neutral personalities, curated role compatibility, and personality invariants. Implementation now lives in [agent-compose#10](https://forgejo.coilysiren.me/coilyco-flight-deck/agent-compose/issues/10). The ownership split is: * Agent-compose embeds Kai public-safe person configuration: organizational purpose, personality definitions, compatibility, selection, and composed context. * Ward retains executable role authority, permissions, credentials, safety gates, and rollback requirements. Shared role slugs are a cross-repo contract, not copied permission policy. * AOS retains reusable doctrine, general skills, capability providers, and editorial validators. * AOSK may add private overlays without becoming the canonical home of the public person source. The autonomy invariant and operating-mode rules in this issue carry forward unchanged. No generic trait-slider or neutral enterprise personality framework is part of v0.1.

coilyco-ops referenced this issue

2026-07-23 06:23:49 +00:00

Author grounded, warm, candid, and playful personality skills #624

coilyco-ops referenced this issue

2026-07-23 06:23:49 +00:00

Author optimistic, curious, bold, and meticulous personality skills #625

coilyco-ops referenced this issue

2026-07-23 06:23:49 +00:00

Author imaginative, diplomatic, reflective, and protective personality skills #626

coilyco-ops referenced this issue

2026-07-23 06:23:49 +00:00

Author energetic, nurturing, charming, and quirky personality skills #627

coilyco-ops commented

2026-07-23 06:24:38 +00:00

Member

Provider work is now tracked in AOS

The earlier ownership move was too broad. Agent-compose owns Kai public-safe catalog bindings, selection, and role compatibility, while AOS still owns the reusable role-neutral personality bodies and their source declaration.

Current implementation map:

agent-compose#42 - approved role compatibility matrix, complete
agent-compose#44 - stable catalog bindings, dispatched separately
agent-compose#21 - remaining favorite-color palette
#622 - AOS personality provider tracker
#623 - skill family and shared invariant
#624 through #627 - four skill-authoring batches
#628 - source declaration, consumer validation, and shipped documentation

Ward remains out of scope.

### Provider work is now tracked in AOS The earlier ownership move was too broad. Agent-compose owns Kai public-safe catalog bindings, selection, and role compatibility, while AOS still owns the reusable role-neutral personality bodies and their source declaration. Current implementation map: * agent-compose#42 - approved role compatibility matrix, complete * agent-compose#44 - stable catalog bindings, dispatched separately * agent-compose#21 - remaining favorite-color palette * #622 - AOS personality provider tracker * #623 - skill family and shared invariant * #624 through #627 - four skill-authoring batches * #628 - source declaration, consumer validation, and shipped documentation Ward remains out of scope.

coilyco-ops referenced this issue

2026-07-23 06:26:08 +00:00

Publish the canonical personality skill provider for agent-compose #622

coilyco-ops referenced this issue

2026-07-23 06:26:08 +00:00

Define the AOS personality skill family and shared invariant #623

coilyco-ops added the

labels