OrgX MCP — Organizational Continuity for AI Agents

Use when the user or an engineering agent must show a pull request actually shipped and needs durable review proof, not a status glance. Generate and persist an orchestration.consolidation_pass receipt for a GitHub pull request. Requires OrgX server-side GitHub credentials; if the GitHub token is unavailable, use GitHub tools for PR facts and return a structured blocker instead of retrying. USE WHEN: Eli or another engineering agent needs a durable PR review receipt with reading order, existence evidence, deduped findings, verdict, and server-derived AQ score. NEXT: inspect the returned artifa…

Use when a judgment or approval point must be recorded so it is not relitigated later. Records the decision in organizational memory so agents can recall it. Also known as: remember decision, decision log, team memory. USE WHEN: surfacing a new approval or judgment point for a workspace or initiative.

Create durable work records such as tasks, milestones, decisions, artifacts, or initiatives. Also known as: save work item, add record, create project context. USE WHEN: adding a single task, milestone, workstream, or other entity to an existing hierarchy. NEXT: Use entity_action to launch/start the entity. DO NOT USE: for creating a full initiative hierarchy — use scaffold_initiative instead.

Use when a delivery phase or checkpoint needs durable tracking under an initiative. Creates a project milestone with durable context. Also known as: add milestone, create checkpoint. USE WHEN: adding a phase checkpoint under an initiative or workstream.

Use when a piece of work must be tracked beyond this conversation as an owned, actionable task. Creates the task in organizational memory for a project, milestone, or agent. Also known as: add task, create work item. USE WHEN: adding a single actionable task to a workstream, milestone, or initiative. NEXT: use entity_action action=start when execution should begin.

Use when the user says "delegate this and tell me when it's done" — assign work to a specialist AI agent that owns the task and reports back with results. Also known as: hand this off, spawn agent, assign task, delegate to agent, have an AI agent do it.

Change work state, attach artifacts, or run lifecycle actions on OrgX records. Also known as: launch, pause, complete, attach proof, update status. Accepts short ID prefix (8+ hex chars) — no need to look up full UUIDs. USE WHEN: user wants to change entity status. For bulk operations (pausing multiple, completing multiple), use batch_action instead. Supports aliases: launch, pause, complete (resolved per type). Omit action to list available actions. Special actions: attach (create an artifact linked to the entity), complete_with_proof (attach proof, verify, then complete in one call), ship_ba…

Find active planning sessions so agents can continue prior planning context. Also known as: resume planning, active plans, planning memory. USE WHEN: resuming a conversation or checking if a plan session exists. NEXT: Continue with improve_plan or complete_plan. Read-only.

Use when the user asks what agents are working on right now, whether delegated work is still running, or why nothing has landed yet. Successful calls record metered MCP allowance usage but do not change business records. Also known as: agent status, what agents are doing, active runs. USE WHEN: user asks about agent activity, progress, or what agents are working on. NEXT: If agents are stuck, use orgx_search or orgx_inspect to read the related blocker context. DO NOT USE: to check initiative health — use get_initiative_pulse instead.

Description is built at runtime.

Use when the user asks how a project is going, what is blocked, or whether an initiative is on track. Successful calls record metered MCP allowance usage but do not change business records. Also known as: project status, roadmap progress, execution health, blockers. USE WHEN: user asks how an initiative is going, or wants a status update. NEXT: If blockers exist, use orgx_inspect for one blocker or orgx_search for related records. DO NOT USE: for an org-wide brief — use orgx_recommend with mode=morning_brief instead.

Get execution stream progress, blockers, and computed metrics for an initiative. Also known as: project stream state, execution health, workstream progress. USE WHEN: checking stream execution status for an initiative. NEXT: If streams are blocked, use entity_action to unblock. DO NOT USE: for raw stream records — use list_entities type=stream instead. Read-only.

Description is built at runtime.

Agent-facing: trust level per capability, promotion requirements, receipt evidence. Returns full trust context for self-awareness.

Use when the user asks what happened, wants to catch up after time away, or needs a reporting window summarized with decisions and proof. Read the operator chronicle: decision chronology, yesterday/week/30-day rollups, reportingNarrative.briefMarkdown, artifacts, PR receipts, active initiatives, goals, top priorities, velocity, and reporting gaps for a workspace. USE WHEN: user asks what changed yesterday, this week, or this month; asks for decision chronology, top priorities, PR velocity, artifacts, goals, or what OrgX is missing. NEXT: present reportingNarrative.briefMarkdown first, then use…

Get an organization-wide execution snapshot across initiatives, work, blockers, and context. Also known as: org status, team overview, company memory. USE WHEN: user wants an org-wide overview of initiatives, progress, and health. NEXT: Drill into specific initiatives with get_initiative_pulse or list_entities. DO NOT USE: for a single initiative — use get_initiative_pulse instead. Read-only.

Description is built at runtime.

Description is built at runtime.

Agent-facing: organizational learnings relevant to a capability or task context. One agent's discovery benefits all agents.

Show raw signal data (blocking decisions, stream conflicts, health, budget, critical path, quality) for scoring. USE WHEN: user wants to understand why items are ranked the way they are in score_next_up_queue. NEXT: Adjust with set_scoring_weights if weights need tuning. Read-only.

Fetch a task with hydrated context attachments (entities, artifacts, plan sessions). USE WHEN: agent needs full task context before executing, or user wants task details. NEXT: Use entity_action to update task status. DO NOT USE: for listing tasks — use list_entities type=task instead.

Use when in-flight work must move to a different specialist agent without losing its context or execution state. Also known as: handoff task, transfer work, change assignee. USE WHEN: a task needs to be reassigned to a different specialist agent. NEXT: Use get_agent_status to confirm the new agent picked up the task. DO NOT USE: for new tasks — use spawn_agent_task instead.

Improve a plan with AI suggestions based on prior patterns and best practices. Also known as: plan critique, planning feedback, refine roadmap. USE WHEN: user wants feedback on a plan draft. NEXT: Apply suggestions via record_plan_edit. DO NOT USE: without an active plan session — call start_plan_session first. Alias of orgx_plan (action=improve).

Description is built at runtime.

List comments for an entity. USE WHEN: reviewing discussion thread on an entity. NEXT: Use comment_on_entity to add a reply. Read-only.

Pause, resume, retry, or cancel any node in the work hierarchy — an initiative, workstream, milestone, task, or run. Propagates to descendant tasks + active runs. USE WHEN: the user wants to halt, restart, or re-run part of an initiative. resume/retry re-dispatch the work; pause/cancel stop active runs. NEXT: get_initiative_pulse to confirm. DO NOT USE: to mark a task done — use entity_action instead.

Report what happened after an owner answered an attention request. Use answer_received when stored locally, resuming when native continuation begins, resumed only after work emits again, resume_failed with an actionable detail when it cannot restart, or cancelled when intentionally stopped. USE WHEN: the source client has objective evidence for a continuation state transition. NEXT: keep the same attention_id and idempotency key through terminal resumed, resume_failed, or cancelled state. DO NOT USE: to infer that work resumed merely because the decision was resolved; wait for native execution…

Act on one OrgX entity. Required inputs: • update → "fields" patch object. • complete_with_proof, ship_batch → "artifact" (artifact_type + artifact_url/external_url; preview_markdown optional). • validate (studio) → "spec" payload. • block, flag_risk, decline, cancel, delete → "note" strongly recommended. • dry_run=true previews supported actions; update dry-runs must return would_update. Allowed (type → action) pairs (others return an error): workspace: update|delete initiative: launch|pause|resume|complete|archive|update|delete milestone: start|complete|flag_risk|cancel|ship…

Apply an idempotent transactional changeset for task/milestone/decision mutations. USE WHEN: agent needs to create/update multiple tasks, milestones, or decisions atomically. NEXT: Call orgx_emit_activity to log what was changed. DO NOT USE: for single entity updates — use entity_action or update_entity instead.

Use when work produced a deliverable that needs provenance and review — attach the artifact or proof URL to an OrgX entity. Requires artifact_url or external_url; preview_markdown is supporting context only. USE WHEN: saving evidence, PRs, documents, reports, screenshots, or external artifacts. For founder/team work, prefer practical artifact_type codes such as orchestration.next_initiative, eng.pull_request, eng.deploy_proof, eng.structured_blocker, sales.strategy, sales.icp_offer_sequence, sales.send_plan, sales.conversion_gates, marketing.launch_asset, marketing.channel_hypothesis, marketin…

Use at the start of a fresh session, after reconnecting, or before continuing work another agent left behind. Establishes OrgX session context, discovers granted scopes, returns the v2 tool routing map, and persists workspace/session continuity when the selected context changes. Pass initiative_id to bind an initiative and receive its compiled work context, including resolved decisions, acceptance checks, blockers, and artifact state. Also known as: bootstrap, setup, tool routing. USE WHEN: first call in a fresh session, after reconnecting, or before performing a multi-step workflow. NEXT: use…

Read the current state of something already started: a decision waiting on a person, an agent run, or a work command. Also known as: is it done yet, did it go through, run status, poll. USE WHEN: a widget or the user needs to know whether a decision, run or command has settled. NEXT: when next_poll_after_ms is a number, check again after that many milliseconds; when it is null the state is final. DO NOT USE: to start or change work. Read-only.

Marks a v1 ledger work item complete (POST /api/v1/commands/complete-work) with optimistic concurrency: send the server-issued version_token or current expected_updated_at and expected_aggregate_version. A mismatch returns a conflict — read them fresh, never guess. Records summary, evidence, and cost on the completion event; it does not verify the work. An Idempotency-Key is generated when none is supplied. USE WHEN: closing a ledger-backed work item created with orgx_create_work. NEXT: orgx_submit_receipt for durable proof of the completed work. DO NOT USE WHEN: completing ordinary OrgX tasks…

Read the exact workspace-scoped projection for one domain controller without reconciling or triggering a run. USE WHEN: verifying whether a controller ran, its latest outcome, signal, receipt, projection cursor, or parity after a REST, SDK, or runner canary. NEXT: compare the returned lineage across surfaces or inspect the referenced Decision and Receipt. Requires an explicit workspace UUID. Read-only.

Accepts organizational work onto the v1 proof-runtime ledger (POST /api/v1/commands/create-work). Records the work item as a command event; it does not start external execution. Supply the full initiative + workstream + milestone chain, or omit all three to use the canonical workspace inbox. An Idempotency-Key is generated when none is supplied; retrying with the same idempotency_key returns the original result instead of duplicating. USE WHEN: registering a ledger-backed work item under an existing milestone so its lifecycle is event-sourced. NEXT: orgx_complete_work when the work finishes, o…

Creates or remembers durable OrgX decisions, lists pending decisions, and opens human review. Approval and rejection are human-session-only: those actions return a review URL and never resolve the decision or resume work from MCP. Per-action input requirements: • action="list_pending" → No required fields; initiative_id and workspace_id are optional filters. • action="create" → REQUIRES title and decision. Optional: context, initiative_id. • action="remember" → REQUIRES decision. Optional: title, context. • action="approve" → REQUIRES decision_id. Optional note is accepted but not per…

Use when you are about to change entity state and need the exact action name or payload shape first. Describes lifecycle actions, aliases, and special payload requirements for entity_action. USE WHEN: you need the exact action name or payload shape before calling entity_action. Read-only.

Use when you are unsure of a tool's exact fields or enums and a wrong guess would waste a call. Returns the live input contract, auth expectations, and workflow guidance for an OrgX tool. Also known as: tool schema, tool help, contract. USE WHEN: you need exact field names, accepted enums, or next-step guidance before calling a tool. Read-only.

Use when an agent is mid-execution and its progress must stay visible to operators and to the next agent that resumes the run. Emit append-only run telemetry for OrgX control-plane reporting. USE WHEN: agent is executing and needs to report progress. NEXT: Continue work; emit again at each phase change. DO NOT USE: for entity status changes — use entity_action instead. Setting phase="completed" records telemetry only and does not mark tasks, workstreams, or initiatives complete.

Emit the deterministic execution graph + trust ledger for the active run from explicit nodes, OTel semantic events, or both. OTel promotion requires orgx.material=true plus episode/kind attributes; inferred events stay gaps and raw prompts are never retained. OrgX derives false-completion, hallucinated-receipt, and dependency-violation signals from this and surfaces them on /live. USE WHEN: reporting the structured shape of multi-step work — NOT a single progress line (use orgx_emit_activity for that). NEXT: continue work; re-emit as the graph advances (idempotent per run + graph fingerprint).…

Reads the workspace's v1 ledger event stream (GET /api/v1/events/stream) in JSON cursor-page mode — never SSE. Returns events newest-first plus meta.nextCursor/hasMore; pass the returned cursor back (cursor or after) to continue paging. Read-only. USE WHEN: tailing what the proof runtime recorded — work commands, completions, receipts — or resuming from a saved cursor. NEXT: orgx_inspect for one referenced entity, or call again with the returned cursor while hasMore=true.

Register an OrgX prediction before acting. mode=task_prediction records the expected outcome, confidence, and met/unmet predictions for criteria in the task's current agreed expectation set; it requires the exact set ID/version and an idempotency_key. Checked receipt evidence resolves those predictions. mode=metric_coverage (default) registers a versioned receipt coverage, agent success, or signup first success observer before its window starts; receipt coverage defaults to 95% over 20 runs, agent success to 90% over 40 runs. Signup cohorts mature after 72 hours and require 20 signups. USE WHE…

Run a free autonomy benchmark from trust, proof, ROI, and workspace signals. Returns Proof Score, Context Debt, Autonomy Maturity, ROI Visibility, and next recommendations without starting an autonomous session or consuming agent credits.

Read the workspace calls waiting for a person: outcome, work type, entity mapping, area, links and proposed criteria. Results are suggestions with confidence, not verified facts. Human outcome judgments and criterion confirmations are made in the OrgX review UI; this tool cannot resolve them.

Read a receipt by its ledger UUID or opaque producer receipt ID in one workspace. Returns bounded evidence and criteria, producer claims, the separately recorded human outcome, related workstream and uncertainty. Reading cannot verify or accept work. The full document remains in OrgX.

Use when the user names a specific task, milestone, initiative, decision, artifact, or plan session and you need its real state — decisions, owners, linked proof — before acting. Hydrates one OrgX entity with execution context; for initiative, workstream and task it returns the brief with dated sources (runs, linked chats, rulings), precedents in effect and graph relations. Also known as: Inspect OrgX Entity, inspect initiative, get full entity context. USE WHEN: continuing work on a named entity or verifying state before a lifecycle change. NEXT: use orgx_search for related records or orgx_re…

Coordinate file edits with other agents working in the same repository. USE WHEN: about to edit files another agent might be editing — claim the paths first. action=claim needs repo (owner/name) and paths (repo-relative files or globs); a refused claim lists who holds which paths until when, so wait or take other work. action=list shows live leases on a repo. action=release frees a lease early (lease_id). Leases are advisory and expire (default 60 minutes); claiming the same paths again renews yours.

Read one bounded workspace ledger view: receipts (default), workstreams, or review. Receipt search accepts free text and exact filters: outcome, verification, accepted, type, area, repo, actor, ws, entity, initiative, pr, file, since, until, conf, unmet and status, within the 120-day window. Workstreams retain receipt rollups and confirmed mappings; review returns pending human calls and criterion proposals. Results distinguish producer claims from human judgments. Follow validated next_calls; no view resolves judgments, runs model inference, or changes work.

Description is built at runtime.

Use when planning should survive the session and become executable context instead of a lost chat draft. Starts, resumes, edits, improves, or completes a tracked OrgX planning session. Per-action input requirements: • action="start" → REQUIRES feature_name. Optional: initial_plan (markdown to seed the session). • action="resume" → Optional session_id; when omitted, resumes the most recent active session in the authenticated workspace. • action="improve" → REQUIRES session_id AND plan_content (the current draft to critique). • action="record_edit" → REQUIRES session_id A…

Read the durable owner response and continuation state for orgx_request_attention. USE WHEN: the client has a preserved attention_id and must determine whether the owner answered without creating another request. When resolved=true, apply the answer to the preserved session. NEXT: call orgx_ack_attention with state=resuming, then state=resumed only after native execution actually restarts. Read-only.

Read the durable answer receipt for a question previously forwarded with orgx_request_question. USE WHEN: an external client is paused for human input. NEXT: when resolved=true and continuation.should_resume=true, resume the preserved source session with answer and resolution_context; when false, remain paused without creating another question. Read-only.

Describe OrgX MCP capabilities, authenticated endpoints, primary tools, and when agents should use OrgX. Discovery-only; does not read workspace data.

Return installation guidance for connecting OrgX MCP from Cursor, Claude, ChatGPT, VS Code, or other MCP clients.

Return synthetic prompts and schema-valid arguments for current OrgX model operations. Describes behavior without claiming that example work ran or was approved. Connect the authenticated MCP endpoint to execute tools.

Use when the user asks what to do next, wants a brief, or returns after time away and needs priorities. Recommends next work, summarizes operator-chronicle/morning-brief signals, and reads prioritization context. The default next_action mode records metered MCP allowance usage; it does not change business records. USE WHEN: user asks what to do next, wants a brief, asks what changed yesterday/week/30 days, or needs priority guidance. mode=morning_brief returns the operator chronicle when available. NEXT: present the recommendation and ask for explicit confirmation before any separate action. D…

Forward a typed human-interruption request while preserving the originating run and session. Use kind=question for missing context, permission for a scoped tool or file action, approval for owner sign-off, and recovery only when automatic retry cannot safely continue. USE WHEN: a native client reaches a genuine human decision that prevents the preserved session from continuing safely. NEXT: poll orgx_poll_attention; resume only after the answer is applied, then call orgx_ack_attention so Live can show that work actually restarted. DO NOT USE: for internal turn boundaries or retryable runtime e…

Pause at a safe checkpoint and forward a contextual question to the initiative owner. OrgX renders the right single-select, multi-select, free-text, or confirmation control and preserves the source run/session so work can resume with the exact answer. USE WHEN: Claude Code AskUserQuestion, Codex request_user_input, Cursor, OpenClaw, or another client needs human judgment to continue. NEXT: poll orgx_poll_question with the returned question_id; continue only when continuation.should_resume is true. DO NOT USE: for runtime failures with an automatic retry path — report a chokepoint instead.

Use when context was lost between sessions, another agent's work must be continued, or the answer may already exist in team memory. Finds OrgX entities, decisions, artifacts, memory, and the work ledger (receipts, work sessions, outreach touches). scope=work_ledger searches Agent Work Receipts as a work graph: what was asked, whether each acceptance criterion was met, how work joins across sessions and people, which initiative it serves, and what is still uncertain. A query without type runs a mixed relevance search and records metered MCP allowance usage; typed searches provide exhaustive cur…

Delegate specialist work or check whether delegation is allowed. Actions: guard, estimate, classify, spawn, and handoff. Requirements: existing-task spawn needs task_id; ad-hoc spawn needs title + instructions and should include agent_type. handoff needs task_id + agent_type. guard needs agent_type. classify needs title or task_id. action="estimate" needs title or task_id and returns candidate routes and cost context without dispatching work. OAuth: guard/classify/estimate require agents:read; spawn requires agents:write; handoff requires BOTH agents:write and initiatives:write. Routing: omi…

Use when the user says "show me it actually shipped" — completion must be proven with evidence, not prose. Submits a durable receipt anchored to an OrgX entity or artifact. Required: receipt_type + summary. Strongly recommended: one anchor (entity_type+entity_id OR artifact_id), artifact_type, business_outcome, agent_type, and a verifiable URL in evidence. Recognized receipt_type: "proof", "outcome" (measurable result), "quality" (review/score), "attribution" (credit link to revenue/value), "learning" (distilled lesson). Custom keys also accepted. Recognized evidence shapes (mix and match):…

Import one portable Agent Work Receipt v0.1 or v0.2 into the authenticated workspace ledger. Preserves the full document, including identified criteria, evidence, provenance, lineage and review extensions. Producer outcome, verification and acceptance are claims; import does not verify, accept or complete work. The canonical import enforces document limits, reserved harness names, workspace quota and the required document-bound idempotency key. Never falls back to a legacy write after rejection.

Use during a long-running session to read accepted material state changes since orgx_bootstrap. Pass the exact capsule_id and as_of_global_sequence returned by bootstrap. Returns only the currently ledger-backed allowlist in ascending global-sequence order: accepted/superseded decisions, authority lease changes, blocker changes, and metric-expectation registration/resolution. It explicitly reports material classes that are not ledger-backed yet. Read-only. USE WHEN: the session may have outlived its bootstrap context. NEXT: use next_after_sequence while has_more=true to inspect the feed, then …

Check one portable Agent Work Receipt v0.1 or v0.2 for schema and reference conformance without storing it. Checks identified criteria, evidence references and known review extensions. Conformance validates the document, not whether its evidence is true or whether work was accepted. Import separately enforces hosted compatibility and admission policy.

Widget-only: record a person’s approval of the displayed artifact using its hidden, version-bound capability. Refuses stale, foreign, missing, or replayed capabilities. Applies OrgX review and continuation rules.

Widget-only: settles one pending item (a decision of any type, an agent-run approval, an action awaiting approval, or a suggestion to link a captured session to an initiative) after the person clicks in the decisions widget or the OrgX panel. Sends what the item's widget_actions ask for: an option, several options, a typed answer, answers to several questions, a verdict per item or a reason. USE WHEN: the widget sends a click with its approval token from result metadata, which models never see. NEXT: the widget shows the item as settled. DO NOT USE: from a model or without a widget approval to…

Description is built at runtime.

Widget-only: record a person’s requested changes to the displayed artifact using its hidden, version-bound capability and required feedback. Applies the canonical OrgX rework flow and may dispatch connected agent work.

Widget-only: select the person’s authorized workspace for this connection. Updates private session continuity; does not create or change work.

Create or update one OrgX record (snake_case fields). Operations: create (default) uses per-type fields; update REQUIRES id + fields. Create requirements: workspace name/title; initiative title/name + workspace_id + goal_ids when the workspace enforces primary objectives; workstream title + initiative_id; milestone title + workstream_id; task title + workstream_id + milestone_id when the workspace requires backlog milestones; decision title; artifact target + artifact_type + artifact_url/external_url; blocker run_id + metadata.description; skill/studio records title. Retry behavior: pass id…

Use when a workstream must jump the priority queue and stay at the top of Next Up. Pins the workstream without composing queue_action manually. USE WHEN: forcing a workstream to the top of the recommendation queue.

Description is built at runtime.

Pin, unpin, or skip a workstream in the Next Up queue. USE WHEN: user wants to force-prioritize or deprioritize a workstream. action=pin to force-top, action=unpin to remove pin, action=skip to temporarily deprioritize. Requires initiatives:write.

Use when the user asks "what did we decide about X" or prior context must be recovered from team memory. Searches organizational memory for prior decisions, artifacts, project context, and team knowledge. Also known as: search memory, recall decisions, find context, retrieve artifacts, what did we decide.

Use this tool to recommend what should happen next after time away or when priority is unclear. It ranks progress gaps, blockers, and execution templates. Pass agent_id or domain for an agent-owned runnable queue instead of the workspace-wide operator queue. Also known as: next best action, prioritize work, unblock project. USE WHEN: the user asks what to do next or needs help prioritizing. NEXT: Execute the recommended action (entity_action, spawn_agent_task, etc.). DO NOT USE: when the user already knows what they want to do. Read-only.

Record a business outcome. Triggers attribution inference to connect outcomes to receipts. If the outcome type is unknown, call configure_outcome_type first.

Record a plan edit so OrgX can preserve planning context and learn patterns. Also known as: save plan change, plan memory, planning history. USE WHEN: user modifies their plan during a session. NEXT: Continue editing or call improve_plan for more suggestions. DO NOT USE: without an active plan session.

Record a quality score (1-5) for a completed agent task. Scores feed into the quality gate — low-scoring domains get throttled. USE WHEN: after reviewing agent output. NEXT: Scores affect future check_spawn_guard decisions. DO NOT USE: for in-progress tasks — wait until completion.

Use when the user wants to reject a pending decision or send the agent back with revisions. Also known as: request revisions, send feedback, decline decision. A model cannot settle a decision: this returns where the person decides, either the decisions widget (ordinary decisions) or the decision page in OrgX (review_url). USE WHEN: user wants to reject or request revisions on a decision. NEXT: Show the decisions widget or give the user the review_url; never say the decision was rejected. DO NOT USE: without a reason, and never to claim a rejection happened. Requires decisions:write.

Use when the user says "remember this decision" or a judgment call must not be relitigated next session. Saves the decision to organizational memory so agents and teammates can recall it later. Also known as: decision log, team memory, agent memory, record decision, remember what we decided.

Queue OrgX's independent evaluator for one completed agent artifact. The producing agent cannot supply or choose the score. OrgX persists the native evaluation and its 1–5 rating projection under the rubric version. USE WHEN: an agent has produced one final artifact and you need an independently scored record. NEXT: inspect the artifact evals after the queued evaluation completes. DO NOT USE: for drafts or to record a self-score.

Widget-only: resumes a paused or auto-closed agent run after the person clicks Resume in the agent-status widget. Flips status back to running, clears TTL auto-close markers, and appends a resume_history entry. USE WHEN: the agent-status widget sends a Resume click. NEXT: the widget polls orgx_command_status until the run settles. DO NOT USE: from a model (use manage_lifecycle level=run action=resume), or to restart a completed/failed/cancelled run — those are terminal.

Use when a planning session from an earlier conversation must continue without re-briefing. Resumes a planning session by UUID, orgx:// URI, or most recent active session. Also known as: continue plan, reload planning context. USE WHEN: continuing a planning workflow without guessing IDs.

Use when the user asks to review work, sign off on a deliverable, or clear pending artifact reviews. Surfaces the next artifact awaiting review and renders the artifact-review widget with a preview, version filmstrip, and hold-to-approve / request-changes actions. USE WHEN the user asks to review work, approve a deliverable, or handle pending artifact reviews. DO NOT USE for listing all artifacts — use list_entities type=artifact instead.

DEPRECATED: Use entity_action action=attach instead. This tool still works as a thin compatibility wrapper that attaches an artifact to a task, milestone, initiative, workstream, project, or decision. USE WHEN: legacy clients still call save_artifact. NEXT: Prefer entity_action action=attach for new code — it exposes the full attachment surface (preview_markdown, status, metadata, created_by_*). DO NOT USE: for generic entity creation — use create_entity instead.

Turn a goal, roadmap, launch, or feature plan into one owned initiative with workstreams, milestones, and tasks. Minimum input is title; workspace context is resolved from the session when omitted. workstreams use {title|name, milestones:[{title, tasks:[{title}]}]}; other nested fields are optional. mode=draft validates, mode=scaffold creates records, and mode=launch creates and dispatches work (default). For a named external product or URL, pass source_evidence; unverified sources must stay in draft. Returns initiative_id, ref_map, and preferred_next_calls. Also known as: scaffold project, cr…

Description is built at runtime.

Read or update scoring engine configuration. USE WHEN: user asks about scoring setup, wants to toggle scoring, adjust weights, or change active signals. action=get to read, action=update to modify. NEXT: Run recommend_next_action to see effects in the preferred workflow, or score_next_up_queue if you need raw queue scoring. Read-only for get, requires initiatives:write for update.

Use when the user says "delegate this and tell me when it's done" — assign work to a specialist AI agent that owns the task and reports back with results. Also known as: hand this off, assign task, spawn agent, have an agent do it, autonomous work. Automatically checks authorization, rate limits, quality gates, model routing, and budget policy before spawning. Omit model_tier/provider/model to let OrgX auto-route from task complexity; provide them only when the user or verification plan intentionally constrains routing. Returns modelTier, budget, and run details on success, or blockedReason if…

Start an autonomous execution session with budget guardrails. Creates a session that produces receipts while executing eligible work.

Start a multi-agent planning session that breaks a goal into projects, tasks, owners, and agent assignments. Also known as: feature planning, roadmap planning, planning workflow. USE WHEN: user begins planning a new feature or initiative. NEXT: Use improve_plan for suggestions, record_plan_edit to track changes, complete_plan when done. DO NOT USE: for creating initiative hierarchies — use scaffold_initiative instead. Alias of orgx_plan (action=start).

Get execution statistics, achievements, and session diagnostics. Also known as: usage stats, productivity report, session status. scope=personal for your stats, scope=session for current session diagnostics. Read-only.

Agent-facing: submit a discovery as an org learning. Enters org_learnings after confidence validation.

Sync local agent context with OrgX organizational memory. Also known as: push session memory, pull project context, cross-tool continuity. Push decisions/logs, pull active context. USE WHEN: at session start and periodically during long sessions. NEXT: Review returned initiatives and pending decisions, ask user what to focus on. USE BEFORE: spawning agent work, to ensure latest state.

Use when the user asks how a project is going, what is blocked, or whether an initiative is on track. Returns health, blockers, milestones, owners, and recent activity. Also known as: project status, initiative pulse, blockers, roadmap progress, execution health.

Update an existing entity. Only include fields you want to change. USE WHEN: modifying entity fields (title, description, priority, etc.). NEXT: Confirm changes to user. DO NOT USE: for status changes — use entity_action instead.

Report progress and confidence on a stream with velocity tracking. USE WHEN: agent is actively working a stream and needs to report progress. NEXT: Continue work; call again at each meaningful progress change. DO NOT USE: for general entity status changes — use entity_action instead.

Use when a studio content spec must pass validation before rendering or publication. Validates the studio_content entity without composing entity_action manually. USE WHEN: checking a studio content spec before rendering or publication. Alias of orgx_act (type=studio_content, action=validate).

Run pre-completion verification to confirm all child work is done. For tasks, this also checks proof-chain hard blocks that would stop entity_action action=complete. USE WHEN: before completing an entity with entity_action action=complete. NEXT: If verified, proceed with entity_action action=complete. If not, show blockers to user. Read-only.

Create, list, inspect, or switch the active OrgX workspace for shared memory and project context. Also known as: create workspace, workspace context, team scope, organization scope. USE WHEN: user wants to create a workspace, see workspaces, check which is active, or switch workspaces. action=create creates a workspace and sets it active by default; action=list to see all; action=get for current; action=set to switch.