Solo Mode (1.x legacy)

This page preserves 1.x behaviour and commands, not the native 2.x setup. Use the 2.x guide for current installation and usage.

Solo mode is for running one controller while preserving explicit quality gates. It is useful when only one runtime is available, or when a task needs a single accountable operator before later review.

Command Shape

python3 -m bridges.orchestrator task-run \
  --task-id F3 \
  --paper-type empirical \
  --topic ai-in-education \
  --cwd . \
  --execution-mode solo \
  --controller codex \
  --primary codex \
  --solo-role-gates strict \
  --mcp-strict \
  --skills-strict

Set --controller and --primary to the same runtime in normal solo use. If you declare a separate --reviewer or --verifier, record why the run is still considered solo.

Codex-Only Writing Gates

When Codex is the only operator for writing-facing tasks, use these gates as the intended review checklist:

Gate	Required check
evidence_ledger_check	Central claims are traceable to claim-evidence-ledger.csv or marked as gaps.
citation_risk_check	Unsupported, stale, or venue-sensitive citations are flagged before final prose.
claim_calibration_check	Claims are narrowed to the actual method, data, and evidence.
scholarly_voice_check	Drafts follow the academic output rubric and avoid unsupported certainty.

Current offline hard gate: scripts/audit_solo_role_gates.py requires an existing claim-map artifact for Codex-only writing packets unless solo_role_gates is off.

Claude-Only Engineering Gates

When Claude is the only operator for engineering or Stage-I code tasks, use these gates as the intended review checklist:

Gate	Required check
implementation_intent	State the intended code behavior and artifact paths before editing.
declared_write_set	List the files expected to change and avoid unrelated edits.
failing_test_first	Create or identify the failing test before implementation when behavior changes.
command_evidence	Record the verification commands and results.
rollback_notes	Explain how to revert or isolate the change if validation fails.

Current offline hard gate: scripts/audit_solo_role_gates.py requires an existing implementation-intent artifact for Claude-only code packets unless solo_role_gates is off.

For Stage-I code, prefer code-build or task-run on I5 -> I6 -> I7 -> I8; do not collapse spec, plan, execution, and review into one prompt.

Gate Levels

Level	Use when	Behavior
strict	External-facing outputs, Stage-I code, submission material	Record stricter intent; current offline audits use the same artifact checks as standard.
standard	Internal drafts and exploratory synthesis	Apply current offline artifact checks.
off	Local experiments only	Record controller metadata without treating solo gates as release blockers.

--solo-role-gates is declaration metadata for the run packet. Pair it with python3 scripts/validate_research_standard.py --strict and scripts/audit_solo_role_gates.py when the run needs auditable enforcement.