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

Closed
opened 2026-07-16 08:15:01 +00:00 by coilyco-ops · 4 comments
Member

Parent

#1

Source decision

The desired personal configuration is specified in agentic-os#602. Agent-compose now starts as a public personal repo rather than a neutral enterprise composition framework. It may embed Kai public-safe person configuration as a first-class product source.

What to build

Commit and embed the opinionated ten-seat synthetic company, role-neutral personality catalog, curated role-personality compatibility graph, and shared personality invariants. A default agent-compose install should be able to resolve Kai profiles from this source without fetching organizational policy from AOS or Ward.

Ownership boundary

Agent-compose owns organizational purpose, personality definitions, compatibility, profile selection, and the context those choices add. Ward owns executable role authority, permissions, credentials, and safety gates. AOS owns reusable doctrine and general skills. AOSK may add private overlays, but the embedded person source remains public-safe. Shared role slugs are a contract, not permission data copied into agent-compose.

Acceptance criteria

  • README and AGENTS position the project as Kai personal, opinionated system published in the open, not a generic enterprise framework.
  • One committed KDL source defines the ten roles: director, advisor, pm, designer, engineer, qa, ops, sales, social, and customer-success.
  • Each role has a documented organizational purpose independent of its Ward permissions.
  • Sixteen role-neutral personalities exist: optimistic, curious, bold, meticulous, grounded, warm, candid, playful, imaginative, diplomatic, reflective, protective, energetic, nurturing, charming, and quirky.
  • Every personality defines positive presence, attention, tempo, and voice without mentioning a particular company role.
  • The curated three-personality compatibility set for every role matches agentic-os#602.
  • Role, personality, model, harness, reasoning effort, permissions, and task acceptance remain separate dimensions.
  • The shared invariant states that personality may affect attention, framing, tempo, voice, and tie-breaking only. It never overrides truthfulness, role obligations, permissions, safety, rollback, or completion.
  • Headless engineer, QA, and ops profiles do not add chatter merely to display personality.
  • Interactive and influential-text roles can express personality strongly while preserving evidence and uncertainty requirements.
  • Agent-compose resolves the embedded person source without network access and records role and personality choices in the decision trace.
  • An invalid role-personality pairing fails with the allowed choices and source provenance.
  • Generated profile projections remain uncommitted. The human-authored person KDL is canonical.
  • Representative tasks show recognizably different attention or voice across two allowed personalities without changing validity or authority.
  • No generic trait-slider or arbitrary personality-composition framework is introduced for v0.1.
  • AOS #602 records the implementation ownership move and links back here.

Blocked by

  • Blocked by #2
  • Blocked by #3

Execution type

HITL for role purpose, personality prose, compatibility, and representative-output review. Schema integration and tests are otherwise AFK.

## Parent #1 ## Source decision The desired personal configuration is specified in [agentic-os#602](https://forgejo.coilysiren.me/coilyco-flight-deck/agentic-os/issues/602). Agent-compose now starts as a public **personal** repo rather than a neutral enterprise composition framework. It may embed Kai public-safe person configuration as a first-class product source. ## What to build Commit and embed the opinionated ten-seat synthetic company, role-neutral personality catalog, curated role-personality compatibility graph, and shared personality invariants. A default agent-compose install should be able to resolve Kai profiles from this source without fetching organizational policy from AOS or Ward. ## Ownership boundary Agent-compose owns organizational purpose, personality definitions, compatibility, profile selection, and the context those choices add. Ward owns executable role authority, permissions, credentials, and safety gates. AOS owns reusable doctrine and general skills. AOSK may add private overlays, but the embedded person source remains public-safe. Shared role slugs are a contract, not permission data copied into agent-compose. ## Acceptance criteria - [ ] README and AGENTS position the project as Kai personal, opinionated system published in the open, not a generic enterprise framework. - [ ] One committed KDL source defines the ten roles: director, advisor, pm, designer, engineer, qa, ops, sales, social, and customer-success. - [ ] Each role has a documented organizational purpose independent of its Ward permissions. - [ ] Sixteen role-neutral personalities exist: optimistic, curious, bold, meticulous, grounded, warm, candid, playful, imaginative, diplomatic, reflective, protective, energetic, nurturing, charming, and quirky. - [ ] Every personality defines positive presence, attention, tempo, and voice without mentioning a particular company role. - [ ] The curated three-personality compatibility set for every role matches agentic-os#602. - [ ] Role, personality, model, harness, reasoning effort, permissions, and task acceptance remain separate dimensions. - [ ] The shared invariant states that personality may affect attention, framing, tempo, voice, and tie-breaking only. It never overrides truthfulness, role obligations, permissions, safety, rollback, or completion. - [ ] Headless engineer, QA, and ops profiles do not add chatter merely to display personality. - [ ] Interactive and influential-text roles can express personality strongly while preserving evidence and uncertainty requirements. - [ ] Agent-compose resolves the embedded person source without network access and records role and personality choices in the decision trace. - [ ] An invalid role-personality pairing fails with the allowed choices and source provenance. - [ ] Generated profile projections remain uncommitted. The human-authored person KDL is canonical. - [ ] Representative tasks show recognizably different attention or voice across two allowed personalities without changing validity or authority. - [ ] No generic trait-slider or arbitrary personality-composition framework is introduced for v0.1. - [ ] AOS #602 records the implementation ownership move and links back here. ## Blocked by - Blocked by #2 - Blocked by #3 ## Execution type HITL for role purpose, personality prose, compatibility, and representative-output review. Schema integration and tests are otherwise AFK.
Author
Member

Roster reconciliation note (b646174): the agent names sketched as comments in agentic-os .ward/roles.kdl are now canonical in this repo's embedded person source, adapted into the reviewed contract shape - agent seats with name and pronouns nested under each role.

Absorbed old-world content: six active roles with their purposes (engineer, director, qa, advisor, ops, pm), twelve named seats (opal/fabled for the claude lineage with she pronouns, terran/solar for the codex lineage with he pronouns), and the social and sales stubs ('saled' typo fixed in our copy; the agentic-os file still carries it). Guardfiles, models, and reasoning effort stay in ward, joined by the shared role slug. Names are opaque strings to the engine.

This issue's remaining scope: assign compatible personality sets to the seven roles beyond engineer, per-seat default personalities, the full sixteen-personality catalog as skills, and reconciling the original ten-seat framing from agentic-os#602 with the twelve-seat reality above.

Filed by Claude Code working the issue queue.

Roster reconciliation note (b646174): the agent names sketched as comments in agentic-os .ward/roles.kdl are now canonical in this repo's embedded person source, adapted into the reviewed contract shape - agent seats with name and pronouns nested under each role. Absorbed old-world content: six active roles with their purposes (engineer, director, qa, advisor, ops, pm), twelve named seats (opal/fabled for the claude lineage with she pronouns, terran/solar for the codex lineage with he pronouns), and the social and sales stubs ('saled' typo fixed in our copy; the agentic-os file still carries it). Guardfiles, models, and reasoning effort stay in ward, joined by the shared role slug. Names are opaque strings to the engine. This issue's remaining scope: assign compatible personality sets to the seven roles beyond engineer, per-seat default personalities, the full sixteen-personality catalog as skills, and reconciling the original ten-seat framing from agentic-os#602 with the twelve-seat reality above. Filed by Claude Code working the issue queue.
Author
Member

Decomposition per the planning discussion (docs/integration.md landed in 4207b4a):

  1. The schema is done - roles, seats, pronouns, personality bindings, and colors all shipped with parse gates. This issue's remaining substance is content.
  2. #21 (HITL) collects the policy inputs: compatibility matrix, per-seat defaults, catalog reconciliation, palette.
  3. Once #21 is filled in, authoring fans out AFK: each personality is an independently verifiable slice - a skill directory that passes the parse gate, composes from a fixture request, and carries a legible color. Batches of 4 keep commits reviewable.

Blocked by #21 only.

Filed by Claude Code working the issue queue.

Decomposition per the planning discussion (docs/integration.md landed in 4207b4a): 1. The schema is done - roles, seats, pronouns, personality bindings, and colors all shipped with parse gates. This issue's remaining substance is content. 2. #21 (HITL) collects the policy inputs: compatibility matrix, per-seat defaults, catalog reconciliation, palette. 3. Once #21 is filled in, authoring fans out AFK: each personality is an independently verifiable slice - a skill directory that passes the parse gate, composes from a fixture request, and carries a legible color. Batches of 4 keep commits reviewable. Blocked by #21 only. Filed by Claude Code working the issue queue.
Author
Member

The personality rollout now has explicit ownership and slices:

  • #42 landed the approved role compatibility matrix.
  • #44 owns the missing stable catalog bindings and load-time cross-reference validation.
  • #21 now contains only the remaining HITL color palette.
  • agentic-os#622 tracks the AOS-owned role-neutral skill provider, decomposed through agentic-os#623-#628.

Seats remain personality-neutral. Ward remains out of scope.

The personality rollout now has explicit ownership and slices: * #42 landed the approved role compatibility matrix. * #44 owns the missing stable catalog bindings and load-time cross-reference validation. * #21 now contains only the remaining HITL color palette. * agentic-os#622 tracks the AOS-owned role-neutral skill provider, decomposed through agentic-os#623-#628. Seats remain personality-neutral. Ward remains out of scope.
Author
Member

The agent audited this umbrella against current main. The embedded source now carries all ten roles with organizational briefings, all sixteen role-neutral personality definitions and colors, the approved three-personality compatibility matrix from #42, stable bindings from #44, the shared invariant, named harness seats, offline resolution, provenance-bearing failures, deterministic traces, and uncommitted projections. ward exec test, build, lint, and concurrent smoke validation pass on the current line. Issue #71 separately owns cross-model representative behavioral evaluation, so that evaluation does not keep the shipped catalog implementation open.

The agent audited this umbrella against current main. The embedded source now carries all ten roles with organizational briefings, all sixteen role-neutral personality definitions and colors, the approved three-personality compatibility matrix from #42, stable bindings from #44, the shared invariant, named harness seats, offline resolution, provenance-bearing failures, deterministic traces, and uncommitted projections. `ward exec test`, build, lint, and concurrent smoke validation pass on the current line. Issue #71 separately owns cross-model representative behavioral evaluation, so that evaluation does not keep the shipped catalog implementation open.
Sign in to join this conversation.
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
coilyco-flight-deck/agent-compose#10
No description provided.