feat: end-to-end task→workflow→outer-loop creation pipeline (#1478)

[colehurwitz]

Closes #1478

Changes

Phase 1: Shared resolve_task() + --task CLI flag

• Added resolve_task() 3-step resolver in factory/task.py: TOML file → Python file → module:Class import string
• Added _load_task_from_python_file() helper using importlib introspection
• Added --task flag to factory outer-loop calibrate and factory outer-loop evaluate commands
• Wired task resolution through SwarmConfig.set_task() for outer-loop integration
• Updated docs/outer-loop.md with Task Discovery documentation
• 12 tests covering all resolution paths, error cases, and equivalence

Phase 2: task-setup workflow mode (W₁₄)

• Added task_setup_workflow() graph definition: fork(researcher_domain, researcher_verification) → join → CEO gate → strategist → user gate → builder → FnNode(validate_task) → archivist (10 nodes, 12 edges)
• Registered in _get_builtin_registry() and WORKFLOW_META
• Added "task-setup" to CEO_MODES in _helpers.py
• Added mode detection line in ceo.md prompt
• SKILL.md auto-generated (gitignored, regenerated by skill_cache)

Phase 3: --task flag for create mode + OptKnob auto-generation

• Added _build_task_aware_directive() in _task_builder.py using same directive-injection pattern as Plugin Package and Update Existing Mode
• Added --task argument to CEO parser in _parser_groups.py
• Threaded task_ref through ceo.py → _ceo_helpers.py → _task_builder.py
• Auto-generates OptKnob mechanical derivation table: AgentNode.role→model, timeout→threshold, prompt_template→prompt
• Added "task-setup" mode suffix handling
• 18 tests covering directive injection, OptKnob tables, resolution failures, and integration

Phase 4: ScoringContract.threshold + domain-level OptKnobs

• Added threshold: float | None = None field to ScoringContract
• Updated TaskDefinition.from_toml() to parse threshold from TOML scoring section
• Extended _build_task_aware_directive() to emit domain-level OptKnob instructions based on threshold presence
• Tests for threshold parsing, default values, and domain knob generation

Phase 5: DataNode OptKnob integration (rebased onto PR #1483)

• Rebased onto PR feat: DataNode graph primitive — first-class data loading in workflows #1483's branch (factory/run-3110c50a) — inherits DataNode, main, and review fixes
• Added ### DataNode OptKnob Auto-Generation section to _build_task_aware_directive()
• DataNode mechanical derivation rules (threshold knobs only):
  • parallelism → threshold knob, bounds [1, current, current*2, 8], expandable=False
  • limit (when set) → threshold knob, bounds [limit//2, limit, limit*2], expandable=True
  • max_items → threshold knob, bounds [100, 250, 500, 1000], expandable=True
• Explicit prohibition of topology knobs for DataNode.split/shuffle
• Each knob uses OptKnob.description field (from PR workflow: make Workflow.to_dict() deterministic; add OptKnob.description #1487)
• 8 new tests for DataNode OptKnob derivation rules

Base branch: factory/run-3110c50a (PR #1483). When PR #1483 merges to main, this PR automatically retargets to main.

Test results

• 1206+ passed (pre-existing k8s/container test failures excluded — no oc/kubectl in CI environment)

🤖 Generated with Claude Code

[github-actions]

Sentrux Quality Report

Absolute

Scanning ....
[scan] git ls-files: 653 total, 639 kept, 14 dropped (ext:14, meta:0, big:0)
[build_project_map] 639 files, 108 unique dirs, 100 cache misses, 4.9ms
[resolve] 1305 resolved, 1684 unresolved (of 2989 total specs)
[resolve_imports] project_map 5.1ms, suffix_idx 1.1ms, suffix_resolve 15.3ms, total 21.4ms
[build_graphs] 639 files | maps 2.5ms, imports 21.6ms, calls+inherit 5.6ms, total 29.8ms | 1304 import, 10576 call, 4 inherit edges
sentrux check — 3 rules checked

Quality: 4414

✗ [Error] max_cc: 9 function(s) exceed max cyclomatic complexity of 30
    factory/workflow/executor.py:_execute_data (cc=53)
    factory/outer_loop/reflector.py:_extract_eval_patterns (cc=46)
    factory/cli/_ceo_helpers.py:_validate_ceo_flags (cc=43)
    factory/cli/_ceo_helpers.py:_execute_ceo (cc=43)
    factory/outer_loop/mutations.py:_try_mutation (cc=40)
    factory/cli/_task_builder.py:_build_ceo_task (cc=34)
    examples/chess_evolve.py:main (cc=33)
    factory/outer_loop/reflector.py:_llm_reflect (cc=32)
    factory/outer_loop/mutations.py:validate_and_repair (cc=31)

✗ 1 violation(s) found

Diff (vs base branch)

Scanning ....
[scan] git ls-files: 653 total, 639 kept, 14 dropped (ext:14, meta:0, big:0)
[build_project_map] 639 files, 108 unique dirs, 100 cache misses, 4.9ms
[resolve] 1305 resolved, 1684 unresolved (of 2989 total specs)
[resolve_imports] project_map 5.1ms, suffix_idx 1.2ms, suffix_resolve 15.2ms, total 21.5ms
[build_graphs] 639 files | maps 2.7ms, imports 21.7ms, calls+inherit 7.9ms, total 32.4ms | 1304 import, 10576 call, 4 inherit edges
sentrux gate — structural regression check

Quality:      4415 -> 4414
Coupling:     0.81 → 0.81
Cycles:       3 → 3
God files:    3 → 3

Distance from Main Sequence: 0.35

✓ No degradation detected