Create a research task or complete its missing brief. Returns its directory and stage guidance. Assemble concrete coverage and expected results from context; brief uses question, audience, scope, output, depth and evidence_standard. Returned clarification_questions guide the agent; do not forward them as a questionnaire. Explain consequential alternatives and resolve critical choices with the user. Choose and pass profile explicitly from the task's purpose and evidence needs; full remains the API default. lite uses compact stage methods, skips review and full delivery gates, and requires its final checklist. Resume an existing task through research_status, which reads its saved profile. Inspect review_readiness before collection: restore blocked dependencies/access; local checks do not verify quota.

Save/update source, claim and requirement objects by stable ID; the tool writes CSV/JSONL. Omitted IDs are assigned or matched by URL/claim/summary. Partial updates retain other fields and rows. draft uses path, optional content and complete=true for a complete draft; omit content to register an existing file. Returns assigned IDs and field errors for the agent to resolve, without asking users to manage records. replace names complete collections to repair from recovered rows; original bytes are backed up automatically. Records reading/verification claims as supplied, never infers truth or closes requirements automatically. Saves observe timestamps and failures; research_status exposes activity and returned reviewer usage.

Read progress and stage methods. Optional stage advances to collect/analyze/draft/revise only after prerequisite records exist. Review and final completion use their dedicated tools.

Read brief/collect/analyze/draft/review/revise/final methods without a task. Pass the intended profile explicitly (API default: full). Lite returns compact guidance for this stage. For an existing task use research_status to inherit its saved profile.

Run the configured content reviews and any declared audits, retaining original evidence. Without a configured backend, new tasks return a self-review prompt; submit the actual result as self_review. Explicit external/independent selections never silently downgrade. Lite returns explicit skip logs. External calls send the full task/report/evidence to the configured model account. May consume its usage. evidence_paths are UTF-8 files relative to the task; required registry materials are included automatically. Use evidence_path in source records for local source text/extracts. Use purpose=evaluation to preserve defects. Resume incomplete assignments; never retry a valid negative result to improve the verdict. After correcting a report, pass revision=True for its new version; this is the normal report-delivery loop. Already authorized corrections do not need another user confirmation. Reconcile changed scope in task records first. For local self-review, include_assignment=False returns the frozen input path/hash instead of repeating its full text. Read the current assignment locally and reuse already-read unchanged evidence; review and freshness checks are identical.

Verify report delivery and publish its receipt/terminal state only when checks pass. Lite returns a checklist template when checklist is omitted; submit each passed/evidence item to finish. Optional evidence_files lists retained task-relative check notes; new Lite receipts also bind research inputs. Explicit length_constraint rows in state/requirements.jsonl are measured for Lite and Full; no limit is inferred from prose. message is the intended user-visible delivery text. This does not send messages or publish externally. For evidenced no-change decisions, append finding_disposition rows to logs/review.jsonl as documented. New ordinary-review attempts require schema_version 2, decision_actor and original artifact/response hashes in those rows. Evaluation tasks end with evaluation_complete from research_review, without repairing samples.

Check draft records and declared whole-artifact length before review or finish, without model calls. Returns actual counts, explicit range checks and errors. No limit is inferred from prose. Does not change progress, reports, reviews or receipts; uses the shared local task lock. ok describes these mechanical checks only, never semantic quality or delivery approval. Correct ordinary drafts and recheck; preserve frozen evaluation originals.

Check trusted reviewer configuration locally, without sending report data or paid model calls.