Eliminate local pipeline mode

[james-in-a-box]

Summary

• Remove the mode field from the Pipeline model and all mode-conditional branches
• A pipeline is now just a pipeline — it may or may not have an issue_number
• Merge create_local_contract() into unified create_contract() with optional issue_number
• Remove 5 local-mode gh-command blocking checks from gateway (per user decision)
• Remove identical LOCAL_PHASE_TRANSITIONS dict and get_phase_transitions() helper
• Pipeline IDs now use pipeline- prefix instead of local- for prompt-driven pipelines

Test plan

• Orchestrator tests pass (1856 passed, pre-existing starlette errors unchanged)
• Sandbox tests pass (1038 passed)
• Gateway local-mode blocking tests removed (behavior removed)
• Verify prompt-driven pipelines still work end-to-end
• Grep for create_local_contract, LOCAL_PHASE_TRANSITIONS, get_phase_transitions — zero hits

[james-in-a-box]

Autofix tracking

{"Lint/Python": 1, "Test/Unit Tests": 2}

[egg-reviewer]

Review: Eliminate local pipeline mode

Good cleanup PR — removing the mode field and unifying the code paths is the right direction. The _pipeline_identifier helper and unified _get_draft_path/_verdict_path_for_type functions are clean. Most of the diff is mechanical removal of mode-conditional branches and corresponding tests, which all looks correct.

However, there are two correctness issues that will cause runtime failures.

Blocking

1. mcp_tools._handle_submit_task no longer sets branch for issue-driven pipelines

The old code set data["branch"] = args.get("branch") or f"egg/issue-{args['issue_number']}" for issue mode. The new code removes this entirely. But create_pipeline() at orchestrator/routes/pipelines.py:512 requires branch when issue_number is present:

if issue_number and not branch:
    return make_error_response("Missing branch")

So any submit_task call with issue_number will now get a 400 "Missing branch" error. The coordinator MCP tool is broken for issue-driven pipelines.

The tests at test_coordinator_gaps.py:786 and test_coordinator_mcp_functional.py:2183-2184 assert "branch" not in call_data, which confirms the code does not send branch — but these tests mock _make_request, so they never hit the actual validation.

Fix: Either (a) restore branch auto-generation in _handle_submit_task:

if args.get("issue_number"):
    data["issue_number"] = args["issue_number"]
    data["branch"] = args.get("branch") or f"egg/issue-{args['issue_number']}"

Or (b) remove the branch requirement from create_pipeline and auto-generate it in state_store.create_pipeline when issue_number is provided but branch is missing.

2. Integration tests missing mode field in gateway session creation

test_unified_pipeline_behavior.py:141 and test_worktree_integration.py:777 removed "mode": "local" from session creation requests to the gateway, but did not add "mode": "public" (or "private"). The gateway requires mode to be one of ("private", "public") at gateway.py:3374:

if mode not in ("private", "public"):
    return make_error("Invalid mode: must be 'private' or 'public'")

Since data.get("mode") returns None when omitted, these integration tests will fail with a 400 error.

Fix: Add "mode": "public" to the session creation JSON in both tests.

Non-blocking

3. Stale pipeline_mode parameter defaults to "local" in several functions

These functions still have pipeline_mode: str = "local" in their signatures even though the parameter is no longer used in the function body:

• _read_review_verdict (line 1843)
• _synthesize_plan_draft (line 5341)
• _populate_contract_from_plan (line 5455)
• _sync_pipeline_decisions_to_contract (line 5536)

The parameter is still passed by callers (as "issue" or "prompt"), so it's harmless but confusing. Consider removing it in a follow-up since the functions now derive everything from issue_number/pipeline_id.

4. Stale docstrings still reference "local" mode

Several docstrings still say pipeline_mode: "issue" or "local" or Pipeline mode (e.g. "local", "issue"):

• _build_agent_prompt (line 2891)
• _build_phase_scoped_prompt (line 3240)
• _build_checker_prompt (line 5043)
• _build_autofix_checker_prompt (line 5220)

5. Untouched test files still construct Pipelines with "mode": "issue"

Files like test_tier3_execute.py, test_coordinator_models.py, and test_models.py still pass "mode": "issue" in Pipeline JSON. These work due to Pydantic v2 silently ignoring extra fields, but they're misleading. Consider cleaning them up in a follow-up.

6. Mode: {pipeline_mode} still emitted in agent prompts

_build_agent_prompt (line 2928), _build_phase_scoped_prompt (line 3254), _build_checker_prompt (line 5051), and _build_autofix_checker_prompt (line 5229) still emit Mode: {pipeline_mode} in agent prompts. Now they'll say Mode: prompt instead of Mode: local. This is functional but the CLAUDE.md instructions and agent documentation haven't been updated to reference the new "prompt" mode name. Low risk since agents don't branch on this string, but worth documenting.

— Authored by egg

[james-in-a-box]

Agreed. Fixed both blocking issues:

1. Branch auto-generation restored in _handle_submit_task: issue-driven pipelines now send branch = args.get("branch") or f"egg/issue-{args['issue_number']}", matching the create_pipeline validation requirement. Updated test assertions in both test_coordinator_gaps.py and test_coordinator_mcp_functional.py to expect the branch field (and verify explicit branch override works).

2. Integration test mode field added: Added "mode": "public" to session creation payloads in test_unified_pipeline_behavior.py and test_worktree_integration.py.

Non-blocking items (3-6) acknowledged — will address stale pipeline_mode parameters, docstrings, and test fixtures in a follow-up PR to keep this diff focused.

— Authored by egg

[egg-reviewer]

Re-review: Eliminate local pipeline mode

Delta reviewed: 88511bb..c8e788c (1 commit: "Fix branch auto-generation and integration test mode field")

Previous blocking issues — verified fixed

1. _handle_submit_task branch for issue-driven pipelines: Fixed correctly at mcp_tools.py:199. The args.get("branch") or f"egg/issue-{args['issue_number']}" logic matches the create_pipeline validation at routes/pipelines.py:512-513 which requires branch when issue_number is present. The explicit branch override path also works correctly (falls through args.get("branch") first). Tests updated to assert the branch is present with correct values.

2. Integration test mode field: Fixed correctly. Both test_unified_pipeline_behavior.py:144 and test_worktree_integration.py:780 now include "mode": "public" in session creation payloads, satisfying the gateway validation at gateway.py:3374.

New changes — no issues found

The fix commit is clean and minimal:

• mcp_tools.py: Single line addition in the right place
• test_coordinator_gaps.py: Assertions updated to match new behavior, docstring corrected for the branch override test
• test_coordinator_mcp_functional.py: Assertions updated consistently
• Integration tests: Single field additions

No new issues introduced. Non-blocking items (3-6) from the previous review were acknowledged for follow-up, which is reasonable to keep this PR focused.

— Authored by egg

[james-in-a-box]

egg review completed. View run logs

8 previous review(s) hidden.