Demo results
The completion demo (examples/mixed-harness, make demo) runs an organisation whose engineering team uses three harnesses on three model endpoints: Claude Code on Anthropic Messages (lead), Codex on OpenAI Responses (engineer, the only seat with GitHub access) and Pi on an OpenAI Chat Completions endpoint (reviewer). The team collaborates without Linear, then with work items and a pull request; work is then published to Linear through an outage; finally the Anthropic key is rotated mid-session. This page is generated by the run.
Runs against fakes and against real services are recorded separately. A fakes run uses a scripted model endpoint, so it proves the wiring (harness, model routing, tools, permissions, recovery), not model quality.
Fakes run, 2026-10-07 23:33 UTC
Pinned versions
| Harness | Version |
|---|
claude-code | 2.1.289 |
codex | 0.160.1 |
pi | 1.0.4 |
Steps
| Step | Result | Time | Evidence |
|---|
| bring up the platform and the mixed organisation | passed | 63s | each seat passed readiness with a probe turn through its own harness and model |
| 1. collaborate through messages and a shared note (no Linear, no work items) | passed | 6s | shared note notes/demo-login.md written by the lead, the engineer and the reviewer: # Login page / lead: plan agreed - engineer builds, reviewer reviews / engineer: building the form and session handling / reviewer: will review the form for accessibility and errors |
| 2. the same with work items; the engineer opens a pull request | passed | 6s | engineering/W-1 owned by the engineer, status done; pull request acme/sandbox#2 (fake GitHub, opened through the platform) |
| 3. publish work items to Linear; an outage never blocks the organisation | passed | 34s | W-1 published once; while Linear was down the organisation stayed ready (IntegrationsDegraded) and W-2 was created internally; after reconnecting W-2 was published, nothing twice |
| 4. rotate the Anthropic key mid-session | passed | 6s | the lead kept working with the rotated key: no Terraform run, no restart (Pod seat-eng-lead-ee1a1cb2-0 unchanged) |
| record the models and endpoints each seat actually used | passed | 6s | 3 seats; see the table |
Models and endpoints actually used
From the model_request events the platform recorded for each seat's runs.
| Seat | Harness | Connection | API | Model | Upstream host | Requests (2xx) |
|---|
eng_lead | claude-code | anthropic | anthropic_messages | claude-sonnet-5-5 | steadmesh-fakes.steadmesh-system.svc:8092 | 23 (23) |
engineer | codex | openai | openai_responses | gpt-5.5 | steadmesh-fakes.steadmesh-system.svc:8092 | 23 (23) |
reviewer | pi | selfhosted | openai_chat | qwen3.8-27b | steadmesh-fakes.steadmesh-system.svc:8092 | 14 (14) |
Configuration (examples/mixed-harness)
{
"access_profiles": {
"github_engineer": {
"github": {
"connection": "github",
"permissions": {
"contents": "write",
"pull_requests": "write"
},
"repos": [
"acme/sandbox"
]
}
}
},
"enable_console": true,
"enable_egress": true,
"enable_linear": false,
"extra_connections": {
"github": {
"adapter": "github",
"endpoint_ref": "http://steadmesh-fakes.steadmesh-system.svc:8093",
"secret_ref": "k8s:github-credentials"
}
},
"harness": "fake",
"model_api_keys": {
"anthropic": "demo-anthropic-1",
"selfhosted": "demo-selfhosted",
"openai": "demo-openai"
},
"model_connections": {
"anthropic": {
"adapter": "anthropic",
"endpoint_ref": "http://steadmesh-fakes.steadmesh-system.svc:8092/v1",
"secret_ref": "k8s:anthropic-credentials"
},
"selfhosted": {
"adapter": "model",
"endpoint_ref": "http://steadmesh-fakes.steadmesh-system.svc:8092/v1",
"model": {
"apis": [
"openai_chat"
],
"models": [
{
"id": "qwen3.8-27b"
}
]
},
"secret_ref": "k8s:selfhosted-credentials"
},
"openai": {
"adapter": "openai",
"endpoint_ref": "http://steadmesh-fakes.steadmesh-system.svc:8092/v1",
"secret_ref": "k8s:openai-credentials"
}
},
"seat_access": {
"engineer": [
"github_engineer"
]
},
"seat_harnesses": {
"eng_lead": {
"adapter": "claude-code",
"model": {
"connection": "anthropic",
"id": "claude-sonnet-5-5"
}
},
"engineer": {
"adapter": "codex",
"model": {
"connection": "openai",
"id": "gpt-5.5",
"settings": {
"reasoning_effort": "medium"
}
}
},
"reviewer": {
"adapter": "pi",
"model": {
"connection": "selfhosted",
"id": "qwen3.8-27b"
}
}
}
}
Real services run, 2026-10-08 00:15 UTC
Pinned versions
| Harness | Version |
|---|
claude-code | 2.1.289 |
codex | 0.160.1 |
pi | 1.0.4 |
Steps
| Step | Result | Time | Evidence |
|---|
| bring up the platform and the mixed organisation | passed | 80s | each seat passed readiness with a probe turn through its own harness and model |
| 1. collaborate through messages and a shared note (no Linear, no work items) | passed | 18s | shared note notes/demo-login.md written by the lead, the engineer and the reviewer: # Demo: login page plan / / Requested by: representative_sean. Coordinator: eng_lead. No work items; this note and messages only. / / ## Goal / A simple login page (username/password form) that validates input and shows clear errors. / / ## Acceptance criteria / - Form with username, password, su… |
| 2. the same with work items; the engineer opens a pull request | passed | 72s | engineering/W-1 owned by the engineer, status in_review; pull request https://github.com/darcys22/flash-fresh/pull/1 |
| 3. publish work items to Linear; an outage never blocks the organisation | passed | 333s | W-1 published once; while Linear was down the organisation stayed ready (IntegrationsDegraded) and W-2 was created internally; after reconnecting W-2 was published, nothing twice |
| 4. rotate the Anthropic key mid-session | skipped | 0s | |
| record the models and endpoints each seat actually used | passed | 6s | 3 seats; see the table |
Models and endpoints actually used
From the model_request events the platform recorded for each seat's runs.
| Seat | Harness | Connection | API | Model | Upstream host | Requests (2xx) |
|---|
eng_lead | claude-code | anthropic | anthropic_messages | claude-sonnet-5-5 | api.anthropic.com | 25 (25) |
engineer | codex | openai | openai_responses | gpt-5.5 | api.openai.com | 41 (41) |
reviewer | pi | selfhosted | openai_chat | qwen3.8-27b | llm.example.com | 24 (24) |
Configuration (examples/mixed-harness)
{
"access_profiles": {
"github_engineer": {
"github": {
"connection": "github",
"delivery": "sandbox",
"permissions": {
"contents": "write",
"pull_requests": "write"
},
"repos": [
"darcys22/flash-fresh"
]
},
"tools": {
"binaries": [
"git",
"gh"
]
}
}
},
"enable_console": true,
"enable_egress": true,
"enable_linear": false,
"extra_connections": {
"github": {
"adapter": "github",
"secret_ref": "k8s:github-credentials"
}
},
"harness": "fake",
"model_connections": {
"anthropic": {
"adapter": "anthropic",
"secret_ref": "k8s:anthropic-credentials"
},
"selfhosted": {
"adapter": "model",
"endpoint_ref": "https://llm.example.com/v1",
"model": {
"apis": [
"openai_chat"
],
"models": [
{
"id": "qwen3.8-27b"
}
]
},
"secret_ref": "k8s:selfhosted-credentials"
},
"openai": {
"adapter": "openai",
"secret_ref": "k8s:openai-credentials"
}
},
"seat_access": {
"engineer": [
"github_engineer"
]
},
"seat_harnesses": {
"eng_lead": {
"adapter": "claude-code",
"model": {
"connection": "anthropic",
"id": "claude-sonnet-5-5"
}
},
"engineer": {
"adapter": "codex",
"model": {
"connection": "openai",
"id": "gpt-5.5"
}
},
"reviewer": {
"adapter": "pi",
"model": {
"connection": "selfhosted",
"id": "qwen3.8-27b"
}
}
}
}