| name | Fryderyk Wiatrowski AI Employee OS |
| description | Design, audit, and upgrade AI employee systems with ambient work surfaces, context isolation, scoped integrations, event semantics, trust ladders, rollout gates, and model/personality regression checks. |
| version | 1.0 |
| format | skill-system |
| workflows | 1 |
| source_url | https://www.youtube.com/watch?v=ohKt066uFhg&t=195s |
| source_package | extractions/video-context/ohKt066uFhg |
| contract | semantic_libraries/antigravity/primitives/ai-employee-operating-contract.md |
Fryderyk Wiatrowski AI Employee OS
Use this skill when the user wants an AI employee, AI coworker, company agent, operating partner, internal operator, or agent system that should behave like a trusted teammate instead of a generic tool.
This skill is source-backed by the Viktor AI coworker talk. It is vendor-neutral for Antigravity use: it does not require Slack, Viktor, Pipedream, or any specific model. It extracts the operating method into a local Codex command system.
Source Grounding
- Evidence package:
extractions/video-context/ohKt066uFhg/
- Evidence map:
extractions/video-context/ohKt066uFhg/evidence-map.md
- Extraction brief:
extractions/video-context/ohKt066uFhg/extraction-brief.md
- Operating contract:
semantic_libraries/antigravity/primitives/ai-employee-operating-contract.md
- Genius context (verbatim exemplars, source-attributed anti-patterns, recognition test):
skills/fryderyk-wiatrowski-ai-employee-os/genius.md
- Claim-by-claim source ledger:
skills/fryderyk-wiatrowski-ai-employee-os/references/source-ledger.md
Command Surface
Run through the workflow:
/ai-employee-os [target system or agent]
/ai-employee-os --audit [target]
/ai-employee-os --design [role/context/tools]
/ai-employee-os --upgrade [route/skill/workflow]
Operating Rule
An AI employee earns scope. Start with role clarity, context isolation, scoped integrations, and trust gates before broad proactivity or autonomy.
Component Order
- Intent lock: decide audit, design, or upgrade.
- Source and target read: load the compact contract, evidence map, and target files only as needed.
- Surface map: identify where the employee lives and which events it receives.
- Context/access map: partition memory and integrations before granting autonomy.
- Employee contract: define job, non-job, inputs, outputs, approvals, and handoffs.
- Proactivity ladder: stage observe, suggest, ask, draft, sandbox act, approved act, and broader activation.
- Regression guard: protect tone, trust, leakage, event semantics, and model swaps.
- Implementation sequence: produce the first safe build or upgrade path.
Supporting Routes
Use these as components, not replacements:
/context-audit for context bloat and compression./memory-architect for persistent memory tiers, decay, and retrieval./conde-agent-experience-design for trust, progress, handoffs, and command UX./24-assets-agent-system-design for roster, ownership, and data flow./source-to-skill-system for future source-to-OS expansions.- Trust architecture and production hardening lenses for permission, security, and regression checks.
Required Output
Every serious run must produce:
- AI Employee OS scorecard
- AI employee system contract
- Context/access map
- Event semantics map
- Proactivity and trust ladder
- Model/personality regression guard
- Validation checklist
- First implementation sequence
Boundaries
- Do not connect Slack, Gmail, Drive, calendar, CRM, or other external systems without explicit approval.
- Do not publish, message, DM, invite, email, or automate accounts unless the user explicitly asks.
- Do not mix client, team, project, or private context without a named access rule.
- Do not treat model replacement as safe because task outputs look correct; check personality and trust.
Quality Bar
- The output names the employee's job and non-job.
- Every context source has a scope and leakage risk.
- Every integration has owner, allowed actions, approvals, and revocation path.
- Ambient events are linearized into a usable event ledger.
- Proactivity is staged, not turned on globally.
- The final path is useful in Codex today and extensible to external team surfaces later.
Execution Prompts (structure-pure v2)
5 deterministic practitioner prompts — each carries an Output Contract, Output Skeleton, and Quality Gate. When a deliverable matches one, Read it and honor its contract instead of improvising the output shape.
- Fryderyk Wiatrowski — AI Employee System Audit —
skills/fryderyk-wiatrowski-ai-employee-os/references/prompts-v2/ai-employee-system-audit.md
- Fryderyk Wiatrowski — AI Employee System Design —
skills/fryderyk-wiatrowski-ai-employee-os/references/prompts-v2/ai-employee-system-design.md
- Fryderyk Wiatrowski — AI Employee System Upgrade —
skills/fryderyk-wiatrowski-ai-employee-os/references/prompts-v2/ai-employee-system-upgrade.md
- Fryderyk Wiatrowski — Integration & Permission Map —
skills/fryderyk-wiatrowski-ai-employee-os/references/prompts-v2/integration-permission-map.md
- Fryderyk Wiatrowski — Model/Personality Regression Canary —
skills/fryderyk-wiatrowski-ai-employee-os/references/prompts-v2/model-personality-regression-canary.md
