Check bridge configuration and target repository connectivity.

Return the target repository branch, commit, dirty status, and top-level entries.

Read a bounded line range from a UTF-8 text file inside the target repository.

Search text files under the target repository without leaving its root.

Start a background Codex turn. EXECUTION_ONLY workstreams require accepted_handoff_path/sha256 and explicit next_workstream: TASK with exact task_id/task_ref or NONE. Omission and legacy next_workstream_task_* inputs are errors. Uses TRUSTED_EXECUTOR_AUTO_ADVANCE with worker-owned DoD and mechanical pointer-only closure. RESEARCH retains candidate/review/RVA acceptance. Without workstreams, completion_policy is MANUAL, BRIDGE_AUTO_MACHINE_VERIFIED, or BRIDGE_AUTO_TRIVIAL; execution-only with zero required obligations defaults to trivial, required obligations to MANUAL.

Return the durable research-cycle artifacts, gate settlements, reviews, and linked finite Codex tasks.

Read the authoritative research graph/ledger snapshot and separately listed migration candidates.

Required post-result step for a research cycle. Call this exact tool after the latest ResultClaim has an acceptable settled mandatory RESULT_REVIEW and before any follow-up decision, next-experiment proposal, cycle closure, or linked execution. task_id MUST be the latest execution task UUID (the latest executionTaskIds entry / ResultClaim.executionTaskId), never the research cycle UUID. Use PUBLISH_AUTHORITATIVE to project the reviewed ResultClaim into deterministic Evidence and publish it directly to the authoritative Knowledge Graph. Use NO_KNOWLEDGE_TO_PUBLISH only when no reusable knowledge exists and provide a non-empty reason. The same action is idempotent, and no separate confirmation is required.

Offline bounded scan of the target repository (default) or bridge-owned data. Extracted artifacts remain PROPOSED.

Atomically persist a text or structured reconstruction/source-registry artifact under an allowed bridge-owned data namespace.

Explicitly confirm or reject one PROPOSED migration candidate with an auditable actor and reason.

Submit a new or revised typed Research Contract without RVA review; the updated contract is persisted and the task is queued for execution.

Backward-compatible alias: record any typed result disposition and queue the generic mandatory result-review gate.

Record a FAVORABLE, NULL, ADVERSE, INCONCLUSIVE, or FEASIBILITY_ONLY result bound to the latest completed execution and canonical artifact hashes, then queue mandatory RVA result review.

After settled result review, explicitly record NO_NEXT_EXPERIMENT or DEFERRED. Formal NEXT_EXPERIMENT_PROPOSED is recorded by rva_review_next_experiment and must settle before closure/linking.

Record the orchestrator's content-based conclusion and next step, bound to the latest settled gate outcome. Required after scientific RVA budget exhaustion before publication, closure, or next execution. Address every objection from both rounds with evidence, preserve unsupported conclusions, and bound any experiment. A decision cannot clear a FAIL_CLOSED provider-failure gate; use rva_retry_unavailable_review after provider recovery.

Explicitly retry the latest RESULT_REVIEW or NEXT_EXPERIMENT_REVIEW after a FAIL_CLOSED RVA provider failure. The original artifact and failure remain auditable. A settlement decision, including BOUNDED_EXPERIMENT, cannot replace a successful RVA review. Scientific review rounds retain their original limit.

Record a next-experiment proposal. Each execution step permits at most two RVA rounds across revisions. The final rejecting round releases the gate immediately while preserving objections; record research_record_settlement_decision before linking execution.

Send a correction immediately to the active Codex turn. awaiting_closure accepts followups and invalidates prior verdict/obligation settlements; completed rejects followups (create a new task). If no turn is active, start a follow-up on the same thread when execution gates permit. Inspect followup.steered or followup.delivery_failed events for delivery confirmation.

Return persisted task plus execution_status, execution_binding (run_id/generation), closure_status, closure_blockers and exact next_workstream TASK/NONE decision (null for legacy/missing/not applicable; never inferred). DRAINED/awaiting_closure consumes no executor slot. COMPLETED means manual or bridge policy closure. AUTO_BLOCKED reports stable verifier reasons; legacy/v1 stays manual without fabricated acceptance.

Read ordered, append-only bridge and Codex events after a sequence cursor.

Return the current repository-wide staged and unstaged diff plus the task baseline reference.

Trusted ChatGPT/bridge governance only. Accept a reviewed agent-workstream-context-v1 CANDIDATE, create its immutable accepted handoff, and atomically advance the configured target repository. Requires a persisted chatgpt.review ACCEPTED JSON binding candidate, prior current SHA-256/revision/handoff, and next task. Task completion/auto-close never authorizes this action. Repeating the same committed transition is idempotent.

Trusted ChatGPT/bridge governance only. final_verdict content: JSON {verdict: ACCEPTED|REJECTED|NEEDS_REVISION, reason?: string}. completion_obligation_settled: JSON {obligation_id: string, evidence_ref: string}. task_closed atomically requires ACCEPTED, settled obligations, drained execution and research gates. bridge.test_evidence machine payload: JSON {version:1, run_id:string, generation:integer, evidence_key:string, status:PASS|FAIL, exit_code:integer}, using current codex_status execution_binding. Both PASS and exit_code=0 are required for command verification. Legacy prose is audit-only. Executors cannot certify evidence or acceptance. Duplicate closure is idempotent.

Read a bounded page of machine-readable audit events and report their research-bridge paths.

Check expected artifact paths and list all currently changed repository paths.

Request cancellation of a queued or running Codex task.