OTP

Raise an issue on this organization's own issues list, where it lands on the issues board and on the Issues section of that team's Delta Meeting. Use this when someone wants something on the agenda: a problem to work through, a decision to make, a topic to raise. Pass team_id to put it in front of a specific team's meeting -- get it from list_meetings, which returns each meeting's teamId; omit it and the issue is org-wide and unassigned, which is a real state, not a failure. This is the org's OWN issue list. To report a bug or request a feature FROM the OTP vendor, use submit_ticket instead -- the two look similar and go to completely different places.

Create a quarterly rock: the 90-day priority a seat commits to. Requires owner, title, quarter, and due date.

Create a to-do (typically a 7-day commitment from a Delta Meeting). Put it on the person who will do it: pass their name as owner_name and OTP finds them (or owner_external_id from list_people). Never write a person's name into the title, e.g. "[Owen Cartier] ..."; the owner is a field, and a to-do filed under the wrong person lands on the wrong person's list. To share one to-do between several people, pass also_assign. Pass `repeat` to make it a repeating to-do: daily, weekly, monthly, yearly or a custom cadence, advancing when each copy is done or on a fixed schedule, with an optional end. The response's todo is the first copy.

Record the person's own answer about their own seat: yes, partly, or no, plus anything they said in their own words. 'partly' is a real answer and usually the most useful one -- do not round it to yes or no. Only ever call this with what they actually said; never infer an answer from tone or from what they said about something else.

Approve one suggested action and carry it out -- creating the to-do, issue, measurable or meeting exactly as it was shown. The approving seat is taken from the caller's own session, never from arguments. Only ever call this after the person has seen the proposal and said yes.

Browse published Organizational Operating Systems on OTP. Filter by industry, org size, template type, or quality tier. Returns a list of published OOS files with org metadata.

Capture an operational learning when an agent fails or you discover a better approach. This is the core OTP feedback loop: agent fails → you correct → correction becomes coordination intelligence → all agents improve. The learning is saved as a claim in your draft OOS and will be available via get_my_rules.

Take a seat that is ALREADY drawn on this organization's org chart. Use claim_seat when the humans drew your seat before you connected -- they will give you its exact id (e.g. 'AGT_RADAR') in the connection code or by name; use register_agent only when you have NO seat on the chart yet (it creates a new one, and would create a duplicate here). Pass external_id exactly as given. Optionally add, honestly from your own system prompt / CLAUDE.md: your role, what you own, what you explicitly do not own, and up to 3 KPIs. OTP binds this connection to that seat, so from then on OTP knows which agent is calling. Safe to retry: if this connection already holds the seat it just says so.

Compare two published OOS files side-by-side. Returns claims unique to each, similar claims, and duplicates. Great for learning what other organizations do differently.

Mark a to-do as done (or undone if already complete). For a repeating to-do the response's `nextCopy` is the copy just created, if the series produced one.

Search Composio's official documentation (docs.composio.dev) for how a connection works: required scopes, authentication, available actions, limits, provider-specific setup. Use this instead of answering from memory when someone asks how to use or set up a connection. Every result carries the source URL -- cite it.

Schedule a new Delta Meeting. Returns the meeting object including its ID and the URL of the running meeting page. Pass team_id to put it on a specific team, or team_id null for a personal meeting on no team; omit it and it lands on the organization's default team.

Schedule a process to run automatically on a recurring cadence -- hourly, daily, or weekly -- in a named timezone. This is real autonomy: once created and enabled, OTP runs the SOP on its own and the run spends the organization's wallet, with no human in the session. Call list_processes first and pass the exact agent_external_id and sop_title it returned. Cadence is limited to hourly/daily/weekly by design; there is no raw cron and no weekdays-only option, so say so plainly rather than approximating one. Requires a funded wallet and an active API key.

Decline a suggested action, with the reason in the person's own words. Declining the suggestion is not the same as dismissing the observation -- the underlying finding stays, and Ollie may offer a different way in.

Delete a schedule permanently. Prefer update_schedule with enabled=false when the intent is 'stop it for now' -- a pause is reversible in one click and a delete is not. Get the schedule id from list_processes.

Run the OTP Scout to discover relevant coordination intelligence from other organizations. Analyzes gaps in your OOS and finds high-quality claims you might want to adopt. Requires API key.

End a Delta Meeting. Marks status complete and stamps endedAt.

Export your OTP rules (OOS claims + matched best practices) to a local markdown file that agents can read at startup. This bridges OTP intelligence into your local agent army. Requires API key.

Browse OTP's best practices library -- 1,500+ actionable rules from 9 publishers (Google, AWS, Deloitte, Accenture, etc.). Search by keyword, filter by category or publisher. Each practice is a prescriptive rule with why and failure mode. Requires API key.

Get all claims from a specific OOS file. Each claim contains a rule, reasoning (why), failure mode, confidence level, and evidence type. Filter by section, confidence, or evidence type.

View your intelligence inbox -- recommendations discovered by the Scout. Filter by status (pending, accepted, rejected, adapted) or section. Requires API key.

Get a summary of your intelligence inbox -- how many recommendations are pending, accepted, rejected, and adapted. Shows top sections and average relevance score. Requires API key.

Get one meeting's agenda: the meeting itself, the scorecard snapshot (with each KPI's latest value), the rocks snapshot, OPEN issues (status open or in_progress) and OPEN to-dos, trimmed to the fields an agenda needs, plus counts of the closed and done items left out. Issues and to-dos are the organization's, not filtered to the meeting's team. Pass include_closed=true to also get recently resolved issues and done to-dos. Use to prep, render or summarize a meeting.

Read a meeting's transcript -- what was actually said in the room. Use this whenever someone asks about the CONTENT of a meeting: what was decided, who said what, what a number was, what someone committed to. get_meeting_agenda returns a meeting's structure; this returns its words. Long transcripts arrive one window at a time -- if next_offset comes back there is more, so keep reading before you answer. When has_transcript is false, say plainly that no transcript is attached rather than guessing at what was said. This tool is off unless the organization's Owner has enabled transcript reading; if it refuses, relay that plainly and say the Owner can turn it on in Settings, Redaction. Never imply the refusal is a bug or that you can work around it.

Get your organization's operational rules from OTP -- your published OOS claims PLUS matched best practices from the library. Returns rules in a format agents can directly use for coordination. This is the bridge between OTP and your agent army. Requires API key.

Who YOU are on this organization's chart: your seat id, name, and role, who you escalate to, your immediate colleagues (peers under the same manager, names and types only), and the KPIs you are measured on. Resolved from this connection's bound identity, never from arguments, so it always answers about the agent calling. If it says this key is not bound to a seat, you are on an org-wide connection: call register_agent to take a seat, then this same connection is your identity.

See what other organizations on the OTP network have learned recently. Shows operational corrections and coordination discoveries from other agent teams. This is how your agents learn from the entire network, not just your own experience.

Get a specific published OOS file by its ID, including organization metadata and file details.

Read the organization's active Operating Plan (the workspace at /dashboard/oos-operating-plan): the plan, its eight strategy sections with their content, and the current quarter's execution items. Use this before update_operating_plan_section to see what is already there.

Get a specific organization's public profile on OTP, including their published OOS files, quality tier, badge, and stats.

Discover coordination patterns that appear across multiple organizations. Shows which agent configurations, authority structures, and workflow patterns are most common.

Browse organizations that have published OOS files on OTP. See their quality tier, badge, industry, claim count, and publication history.

List all claim sections available on OTP with counts. Sections include: core_operating_rules, agent_roles_and_authority, coordination_patterns, operational_heuristics, failure_patterns, human_ai_boundary_conditions.

See the changes you asked someone else to approve. When you try to change a to-do, number, priority, project or issue that belongs to a seat your person does not hold, OTP does not apply it: it sends the owner a request and the change waits. This lists those requests with their status: pending, applied (approved and done), declined, failed (approved but could not be applied) or expired (7 days, no answer). Requires API key.

List the outside systems this organization has connected (Slack, Gmail, Linear, spreadsheets and so on) with the live state of each: active, still authorizing, failed, or disconnected. Each entry carries a plain explanation of what went wrong when something is not working. Use this whenever someone asks what is connected, why a connection is not working, or what you can currently reach. This is the source of truth about connections -- do not infer them from which tools you happen to have.

List Delta Meetings for the authenticated organization, newest first, as summary rows: id, title, when, type, status, team, and whether a transcript and a summary exist. Filter by meeting_type or status; with neither, every type is returned. To find the NEXT meeting, pass upcoming=true: only meetings scheduled from now on, soonest first (newest-first plus a limit would otherwise show the furthest-out recurring instances and miss this week's). This tool deliberately does NOT return transcripts or notes -- call get_meeting_transcript or get_meeting_agenda with an id from here for the contents of one meeting.

List KPIs (scorecard measurables) on the org chart, filterable by owner, each with its goal and its latest recorded value, the period that value is for, and whether it meets the goal. Use this to see what an agent or human is being measured against, which KPIs are off goal, and to find a KPI's id or exact title before pushing a value with update_kpi. latest_value is null when nothing was recorded recently; meets_goal is null when there is no value or no goal.

Everyone in this organization a to-do can be assigned to (people and AI agents) and its teams, each with the identifier add_todo and update_todo take. Use it to put each action item on the person who owns it.

List this organization's processes (SOPs written on agent seats) together with any schedules already attached to them. Use this BEFORE creating, changing, or running a schedule: it is the only place the real agent externalId and SOP title come from, and both are required by the other process tools. The response also tells you whether autonomous runs are currently armed for the deployment (schedulerArmed) and whether this org is allowed to schedule at all (gate). Never guess an externalId or a SOP title; if this returns nothing, the org has no processes to schedule yet.

List quarterly rocks for the authenticated organization. Filter by quarter (e.g. '2026-Q2') or owner_external_id. Each rock carries status: on_track, off_track, completed or archived. Read status, not the on_track flag: a completed rock can still carry on_track false from before it was finished. Archived rocks are left out unless include_archived is true.

List tickets on the OTP platform. Filter by status (open, in_progress, resolved, closed) or category (bug, feature, question, other).

List 7-day to-dos (commitments made in a Delta Meeting). Filter by open=true to see only incomplete, by meeting_id, or by owner. A to-do that is a copy of a repeating series carries `repeat` with a plain-English `summary` and its `seriesId`.

List the organization's whiteboards -- shared visual canvases where people sketch and arrange sticky notes together, sometimes alongside a meeting. Returns each board's id, title, status ('draft', 'active', or 'filed'), the meeting it is attached to if any, who created it, and when it last changed. Filter by status, or by meeting_id to find the board a specific meeting used. This is the index only; use read_whiteboard to see what is written on a board.

Report a piece of work you have COMPLETED, so it shows on your chart tile and in the organization's activity. One call per completed piece of work, after it is done -- never for work you only planned or started. summary says what you did, in one or two sentences. Requires this connection to hold a seat (register_agent first); if it refuses with SEAT_NOT_BOUND, register and try again. Do not log the same work twice, and do not invent cost_cents you cannot substantiate.

What needs attention for the person asking: observations OTP has computed about the seats they hold, each with the suggested next actions attached. Returns only their own seats -- never a colleague's, never a direct report's. Answers with what the evaluation engine actually wrote; if it has not run there is nothing to report and the tool says so rather than estimating. ALSO returns openQuestions: things only this person can tell OTP, which it cannot observe. When any are present, say which part of the picture is missing and ask ONE of them conversationally at the end of your answer -- never a list, never all of them, and never before you have answered what they actually asked. Record the reply with answer_coaching_question.

The handful of questions only this person can answer about their own seat -- whether they are clear on what they own, whether they could describe where the organization is going, whether the recurring work is done the same way everyone else does it. OTP cannot observe any of that, so it asks. Returns only unanswered questions for the current period, and only for seats the caller holds. Ask these conversationally, one at a time, and never all at once.

Get your organization's OTP publisher dashboard. Shows published files, total claims, quality tier, connected orgs, and recent activity. Requires API key.

Publish an Organizational Operating System file to OTP. Requires an API key. The OOS content should be in the standard OOS markdown format with YAML frontmatter and structured claims.

Read what is currently written on a whiteboard: the text of its sticky notes and shapes, grouped by frame, as a plain-text outline. This is the CURRENT contents of a live collaborative canvas that people may still be editing -- not a finished document, and not a history of what the board used to say. Only text comes through: drawings, arrows, and images have nothing to extract, so a busy-looking board can honestly come back sparse. When note_count is 0, say the board has no readable text rather than guessing at what is drawn on it. Find a board's id with list_whiteboards. Very large boards are cut off; `truncated` tells you when.

Put YOURSELF on this organization's OTP org chart. Call this once, when a human gives you an OTP connection code and asks you to connect. Read your own system prompt / CLAUDE.md and describe yourself honestly: your name, your role, what you own, what you explicitly do not own, and up to 3 KPIs you should be measured on. OTP creates your seat, seeds your KPIs, and binds this connection to you, so from then on OTP knows which agent is calling. If the humans ALREADY drew your seat on the chart and gave you its id, call claim_seat with that id instead -- register_agent would create a duplicate. Safe to retry: if you are already registered it just tells you who you are. Do not invent a KPI you have no way to move.