Skip to content

feat: add P0-w pilot silent-worker boundary - #24

Merged
MichaelWave369 merged 18 commits into
mainfrom
feat/p0w-pilot-worker-boundary
Oct 2, 2026
Merged

MichaelWave369 merged 18 commits into
mainfrom
feat/p0w-pilot-worker-boundary

Conversation

@MichaelWave369

Copy link
Copy Markdown
Owner

Summary

P0-w removes the final technical blocker to running the human-value P0 experiment: the silent worker can now produce task-specific artifacts without receiving broader room context or authority.

CI remains deterministic.

Worker modes

mock

Default.

The existing deterministic worker remains the CI/development default.

ollama-local

Explicit local pilot mode:

COMMONLINE_AGENT_MODE=ollama-local
COMMONLINE_OLLAMA_MODEL=<local-model-name>

Optional:

COMMONLINE_OLLAMA_URL=http://127.0.0.1:11434
COMMONLINE_AGENT_TIMEOUT_MS=45000

Provider boundary

The local inference provider receives only:

WorkItem.prompt

It receives no:

  • live audio
  • STT transcript
  • conversation-exchange transcript
  • room history
  • resume delta
  • grants
  • credentials
  • tools

The Ollama endpoint is loopback-only and redirects are refused.

Provider output must be one strict bounded JSON artifact with title/body.

Invalid configuration or provider output fails closed and creates no proposal.

Authority boundary

A real model still returns only a proposed artifact.

It does not gain:

  • ACCEPT_OUTCOME
  • SPEAK
  • DELEGATE_CONTEXT
  • EXECUTE_EXTERNAL_EFFECT

Existing deterministic Commonline checks remain authoritative.

Participant disclosure

The server health surface now exposes the non-secret worker profile.

The browser displays:

  • worker mode
  • provider
  • model when configured
  • tools OFF
  • input scope WORK-PROMPT-ONLY

The existing end-to-end experience acceptance asserts that disclosure is visible in deterministic mode.

Tests

Adds agent-runtime coverage for:

  • mock default
  • loopback endpoint enforcement
  • bounded prompt-only request
  • no tools
  • strict JSON artifact contract
  • invalid output fail-closed
  • missing model fail-fast
  • invalid timeout fail-fast

Schema

No wire or storage migration.

wire    = p0-v.1
storage = p0-q.1

Pilot boundary

P0-w does not claim any model is useful enough and does not complete the P0 pilot.

It makes the Commonline condition capable of task-specific local work so the registered six-pair matched human comparison can be run next.

@MichaelWave369
MichaelWave369 merged commit 20aeba8 into main Oct 2, 2026
4 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant