Remove fixture project agent runtime

.agent-loop/LOOP_STATE.md

@@ -4,17 +4,18 @@
 
 - Active initiative: `WS-POL-001` - Submission Artifact Policy Foundation
 - Active planning chunk: none
-- Active implementation chunk: none
-- Branch: `main`
-- Status: `WS-POL-001-08` is merged to `main` through PR #69 as `aea7024`.
-  Automatic pre-submit project setup now runs through Celery from guide/source
-  capture: sufficiency analysis runs first, blocked sufficiency stops policy
-  derivation, and passed or warning sufficiency creates only a draft
-  `SubmissionArtifactPolicy` for human Workstream approval.
+- Active implementation chunk: `WS-POL-001-09` - OpenAI Agents SDK Only Project
+  Setup Runtime
+- Branch: `codex/ws-pol-001-09-openai-agent-sdk-only`
+- Status: `WS-POL-001-09` is in implementation and internal review. This chunk
+  removes the production `local_fixture` project setup runtime and the old
+  runtime selector so automatic project setup only uses the OpenAI Agents SDK
+  runtime behind the Workstream project-agent port.
 - Last merged implementation SHA: `0c32c97`
 - Last merge commit: `aea7024`
-- Current gate: post-merge memory update for `WS-POL-001-08`
-- Next chunk: inactive until the user gives an explicit start signal
+- Current gate: deterministic verification and internal reviewer fanout for
+  `WS-POL-001-09`
+- Next chunk: inactive until `WS-POL-001-09` reaches a human checkpoint
 
 ## Operating Rule
 
@@ -37,6 +38,11 @@ through Celery. It did not redesign post-submit policy, review, revision,
 payment, reputation, blockchain integrations, frontend behavior, or task
 submission runtime.
 
+The active `WS-POL-001-09` chunk is a corrective hardening chunk for the project
+setup runtime boundary. It removes the production fixture adapter and does not
+change task, checker, post-submit, review, revision, payment, reputation,
+blockchain, frontend, or object-storage behavior.
+
 ## Last Review State
 
 - Last completed initiative: `WS-ENG-001` Codex zero-trust engineering loop bootstrap.

.agent-loop/initiatives/WS-POL-001-submission-artifact-policy-foundation/CHUNK_MAP.md

@@ -665,3 +665,87 @@ Human review focus:
 
 Task contract cleanup without weakening project-policy driven submission
 intake.
+
+### WS-POL-001-09: OpenAI Agents SDK Only Project Setup Runtime
+
+Goal:
+
+Remove the production `local_fixture` project setup runtime and the runtime
+selector that made fixture-derived output look like real agent SDK output.
+Automatic project guide sufficiency and submission artifact policy derivation
+now run only through the OpenAI Agents SDK runtime behind Workstream's
+project-agent port.
+
+Risk:
+
+L1
+
+Depends on:
+
+`WS-POL-001-08`
+
+Allowed files:
+
+```text
+backend/app/core/config.py
+backend/app/adapters/project_agents/**
+backend/tests/test_projects.py
+backend/tests/test_config.py
+README.md
+examples/terminal_benchmark/**
+.agent-loop/LOOP_STATE.md
+.agent-loop/initiatives/WS-POL-001-submission-artifact-policy-foundation/STATUS.md
+.agent-loop/initiatives/WS-POL-001-submission-artifact-policy-foundation/CHUNK_MAP.md
+.agent-loop/initiatives/WS-POL-001-submission-artifact-policy-foundation/chunks/WS-POL-001-09-openai-agent-sdk-only-project-setup.md
+.agent-loop/initiatives/WS-POL-001-submission-artifact-policy-foundation/reviews/WS-POL-001-09-*
+```
+
+Not allowed:
+
+```text
+backend/app/modules/tasks/**
+backend/app/modules/checkers/**
+backend/app/modules/projects/** except project-agent adapter imports if needed
+backend/alembic/**
+.github/workflows/**
+demos/**
+frontend/**
+payment/reputation/blockchain code
+object-storage implementation
+new agent runtime provider implementation
+production secrets or committed .env files
+```
+
+Acceptance criteria:
+
+- Production project setup has no `local_fixture` runtime adapter.
+- `WORKSTREAM_PROJECT_AGENT_RUNTIME_ADAPTER` is removed from active code and
+  current operator docs.
+- The project-agent runtime factory builds the OpenAI Agents SDK runtime and
+  fails closed when the required model setting is missing.
+- Setting the old runtime selector to `local_fixture` does not re-enable a
+  fixture runtime.
+- Tests use explicit test-local fakes for deterministic project-agent behavior.
+- Terminal Benchmark example docs and script do not describe a removed runtime
+  selector or fallback.
+- README explains that automatic project setup needs OpenAI Agents SDK model and
+  API-key settings.
+- Temporary Week 1 demo startup does not enable setup autostart without the
+  required OpenAI worker configuration.
+
+Verification:
+
+- Ruff, focused project-agent tests, docstring coverage, stale wording scan,
+  Markdown link check, and diff whitespace checks pass.
+
+Required reviewers:
+
+senior engineering, QA/test, security/auth, product/ops, architecture, docs,
+reuse/dedup, test delta.
+
+Human review focus:
+
+The removed fixture path cannot be used as a production project setup runtime;
+test fakes stay visibly test-local; operator docs do not imply fallback
+behavior; and the project-agent port remains available for deliberate future
+runtime providers.

.agent-loop/initiatives/WS-POL-001-submission-artifact-policy-foundation/STATUS.md

@@ -4,16 +4,14 @@
 
 `WS-POL-001-01`, `WS-POL-001-02`, `WS-POL-001-03`, `WS-POL-001-04`,
 `WS-POL-001-05`, `WS-POL-001-06`, `WS-POL-001-07`, and `WS-POL-001-08` are
-merged to `main`. `WS-POL-001-08` merged through PR #69 as `aea7024` after
-deterministic verification, internal reviewer fanout, GitHub Actions,
-CodeRabbit review, external review response, and user merge approval.
-
-No implementation chunk is active. The next chunk must start only after an
-explicit user signal and a new chunk contract.
+merged to `main`. `WS-POL-001-09` is active on
+`codex/ws-pol-001-09-openai-agent-sdk-only` after the user's explicit correction
+to remove the production fixture runtime and keep project setup on the OpenAI
+Agents SDK runtime.
 
 ## Active Chunk
 
-None.
+`WS-POL-001-09` - OpenAI Agents SDK Only Project Setup Runtime.
 
 ## Chunk Status
 
@@ -27,12 +25,13 @@ None.
 | `WS-POL-001-06` | Merged | `codex/ws-pol-001-06-terminal-benchmark-drill` | 67 | Hardens the Terminal Benchmark proof harness and removes stale project guide/payment contracts before continuing pre-submit checker work. |
 | `WS-POL-001-07` | Merged | `codex/ws-pol-001-07-task-contract-cleanup` | 68 | Removes task-owned `required_files`/`required_evidence` from request/response/model/migration and keeps artifact requirements project-policy driven. |
 | `WS-POL-001-08` | Merged | `codex/ws-pol-001-08-celery-project-setup` | 69 | Makes guide/source capture enqueue Celery pre-submit setup automatically: sufficiency first, blocked stops, draft submission artifact policy next; removes remaining construction-state compatibility surfaces. |
+| `WS-POL-001-09` | In review | `codex/ws-pol-001-09-openai-agent-sdk-only` | - | Removes the production `local_fixture` project setup runtime and old runtime selector; keeps deterministic test behavior in explicit test-local fakes only. |
 
 ## Blockers
 
 | Blocker | Owner | Next action |
 |---|---|---|
-| None | - | Await explicit user start signal for the next bounded chunk. |
+| None | - | Complete deterministic verification and internal review for `WS-POL-001-09`. |
 
 ## Follow-Ups
 

.agent-loop/initiatives/WS-POL-001-submission-artifact-policy-foundation/chunks/WS-POL-001-09-openai-agent-sdk-only-project-setup.md

@@ -0,0 +1,107 @@
+# Chunk Contract: WS-POL-001-09 - OpenAI Agents SDK Only Project Setup Runtime
+
+## Initiative
+
+WS-POL-001 - Submission Artifact Policy Foundation
+
+## Goal
+
+Remove the production `local_fixture` project setup runtime and the runtime
+selector that made fixture-derived policies look like real agent SDK output.
+Project guide sufficiency and submission artifact policy derivation now run
+through the OpenAI Agents SDK runtime. Tests may use explicit test-local fakes,
+but production code must not expose a no-network fixture adapter.
+
+## Why This Chunk Exists
+
+The Terminal Benchmark live API drill proved the lifecycle plumbing, but it also
+showed that a fixture runtime can confuse the real product question: whether the
+agent SDK can derive a project-specific `SubmissionArtifactPolicy` from the full
+project guide. Workstream should fail closed when the real agent runtime is not
+configured instead of silently producing generic fixture policy.
+
+## Risk Class
+
+L1
+
+## Allowed Files
+
+```text
+backend/app/core/config.py
+backend/app/adapters/project_agents/**
+backend/tests/test_projects.py
+backend/tests/test_config.py
+README.md
+examples/terminal_benchmark/**
+.agent-loop/LOOP_STATE.md
+.agent-loop/initiatives/WS-POL-001-submission-artifact-policy-foundation/STATUS.md
+.agent-loop/initiatives/WS-POL-001-submission-artifact-policy-foundation/CHUNK_MAP.md
+.agent-loop/initiatives/WS-POL-001-submission-artifact-policy-foundation/chunks/WS-POL-001-09-openai-agent-sdk-only-project-setup.md
+.agent-loop/initiatives/WS-POL-001-submission-artifact-policy-foundation/reviews/WS-POL-001-09-*
+```
+
+## Not Allowed
+
+```text
+backend/app/modules/tasks/**
+backend/app/modules/checkers/**
+backend/app/modules/projects/** except project-agent adapter imports if needed
+backend/alembic/**
+.github/workflows/**
+demos/** except README command wording if directly impacted
+frontend/**
+payment/reputation/blockchain code
+object-storage implementation
+new agent runtime provider implementation
+production secrets or committed .env files
+```
+
+## Acceptance Criteria
+
+- Production project setup has no `local_fixture` runtime adapter.
+- `WORKSTREAM_PROJECT_AGENT_RUNTIME_ADAPTER` is removed from active code and
+  current operator docs.
+- The project-agent runtime factory builds the OpenAI Agents SDK runtime and
+  fails closed when the required model setting is missing.
+- Tests use explicit test-local fakes for deterministic project-agent behavior;
+  no production fixture adapter is imported by test setup routes.
+- Terminal Benchmark example docs and script no longer refer to the removed
+  runtime selector.
+- README explains that automatic project setup needs OpenAI Agents SDK model and
+  API-key settings.
+- Temporary Week 1 demo startup does not enable setup autostart without the
+  required OpenAI worker configuration.
+- Stale wording scan, Markdown link check, ruff, focused project-agent tests,
+  docstring coverage, and diff whitespace checks pass.
+
+## Verification Commands
+
+```bash
+cd backend && .venv/bin/python -m ruff check app/core/config.py app/adapters/project_agents tests/test_projects.py tests/test_config.py
+cd backend && .venv/bin/python -m pytest tests/test_config.py tests/test_projects.py -k 'create_guide_autostart_runs_celery_pipeline_to_draft_policy or create_guide_autostart_stops_before_derivation_when_sufficiency_blocks or project_agent_factory_requires_openai_agent_sdk_model or sufficiency_agent_route_is_async_idempotent_and_secret_safe or openai_runtime_misconfiguration_is_sanitized_and_agent_route_only or derivation_agent_allows_warning_report_without_acknowledgement_and_is_idempotent or agent_derived_warning_policy_requires_acknowledgement_before_approval' -q
+cd backend && .venv/bin/python -m pytest tests/test_projects.py -k 'agent or autostart or source_snapshot_integrity or openai or project_setup_pipeline' -q
+cd backend && .venv/bin/docstr-coverage app/adapters/project_agents app/core/config.py --config .docstr.yaml
+python3 scripts/check_stale_workstream_wording.py
+python3 scripts/check_markdown_links.py
+git diff --check
+```
+
+## Required Internal Reviewers
+
+- senior engineering
+- QA/test
+- security/auth
+- product/ops
+- architecture
+- docs
+- reuse/dedup
+- test delta
+
+## Human Review Focus
+
+- The removed fixture path cannot be used as a production project setup runtime.
+- Test fakes are visibly test-local and cannot be mistaken for agent SDK output.
+- Operator docs do not mention a removed runtime selector or imply fallback
+  behavior.
+- Workstream still keeps the project-agent port so future runtimes can be added
+  deliberately in a separate chunk.

.agent-loop/initiatives/WS-POL-001-submission-artifact-policy-foundation/reviews/WS-POL-001-09-external-review-response.md

@@ -0,0 +1,39 @@
+# External Review Response: WS-POL-001-09
+
+## Scope
+
+External review feedback for PR #71: `Remove fixture project agent runtime`.
+
+Internal sub-agent evidence is tracked separately in:
+
+- `.agent-loop/initiatives/WS-POL-001-submission-artifact-policy-foundation/reviews/WS-POL-001-09-internal-review-evidence.md`
+
+## CodeRabbit Review
+
+CodeRabbit completed review on 2026-07-06 and reported one nitpick.
+
+| Source | Finding | Severity | Status | Response |
+|---|---|---:|---|---|
+| CodeRabbit | Repeated deterministic project-agent runtime monkeypatch boilerplate in `backend/tests/test_projects.py`. | Trivial | Fixed | Added the test-local `deterministic_project_agent_runtime` fixture and replaced repeated deterministic monkeypatch blocks. Custom failing/spoofing/capturing runtime patches remain local because they test different behaviors. |
+
+## Verification
+
+```bash
+cd backend && .venv/bin/python -m ruff check app/core/config.py app/adapters/project_agents tests/test_projects.py tests/test_config.py
+cd backend && .venv/bin/python -m pytest tests/test_config.py tests/test_projects.py -k 'create_guide_autostart_runs_celery_pipeline_to_draft_policy or create_guide_autostart_stops_before_derivation_when_sufficiency_blocks or project_agent_factory_requires_openai_agent_sdk_model or project_agent_factory_ignores_removed_runtime_selector or sufficiency_agent_route_is_async_idempotent_and_secret_safe or source_snapshot_integrity_accepts_v1_manifest_without_content_excerpt or sufficiency_agent_blocks_thin_guides or derivation_agent_allows_warning_report_without_acknowledgement_and_is_idempotent or agent_derived_warning_policy_requires_acknowledgement_before_approval or derivation_agent_validates_existing_policy_integrity_before_reuse or agent_derived_submission_artifact_policy_body_is_immutable' -q
+python3 scripts/check_stale_workstream_wording.py
+python3 scripts/check_markdown_links.py
+git diff --check
+```
+
+Results:
+
+- ruff: passed.
+- Focused post-CodeRabbit pytest slice: 11 passed, 188 deselected.
+- Stale wording scan: passed.
+- Markdown link check: passed for 9 changed Markdown files.
+- Diff whitespace check: passed.
+
+## Remaining External State
+
+GitHub Actions and CodeRabbit must rerun after this response and fix are pushed.