Schedules reference

The declarative reference for schedules: the trigger schema, the older standalone schedules block, the self-MCP tools Claude uses to manage a project’s schedules, and the REST surface behind the Triggers tab.

The trigger schema (schedule)

A schedule is a trigger persisted in a project’s project.yaml under the triggers map, keyed by name. Each trigger is trigger (when) + run (what) + enabled:

project.yaml

triggers:
  morning-triage:
    trigger:                       # WHEN — the schedule
      type: schedule
      cron: "0 9 * * *"            # 5-field cron (or @daily / @hourly), host-local
      # interval: "30m"            # …or a duration; exactly ONE of cron / interval
    run:                           # WHAT — the fired agent turn
      prompt: Triage overnight issues and post a summary.
      # promptFile: triage.md      # …or a .md file under .paddock/triggers/; exactly ONE
      session: new                 # "new" (default) | "resume"
      tools: [Read, Grep]          # allow-list = capability; [] = the project's own agent
      model: ""                    # optional model override
      permissionMode: acceptEdits  # default | acceptEdits | bypassPermissions | plan
      maxTurns: 30                 # optional cap (default 30)
      maxSpawnDepth: 0             # optional; 0 = may not spawn children
    enabled: true

The fields:

Field	Values	Notes
trigger.type	schedule	The discriminant. Triggers can also be event (an onArchive / afterTurn lifecycle event) or webhook — the latter is shape-reserved but not yet fireable (no inbound ingress yet). The afterTurn event is reserved for the built-in sweeper/curator.
trigger.cron	5-field string	e.g. 0 9 * * *; @daily / @hourly accepted. Host-local time.
trigger.interval	duration string	e.g. 30m, 1h, 15m.
run.prompt	string	The instruction the firing runs.
run.promptFile	*.md name	Read fresh at firing from .paddock/triggers/; traversal and non-.md are rejected.
run.session	new | resume	new (default) = a fresh chat each firing; resume = one owned accreting session.
run.tools	string array	The fired agent’s allow-list. Empty (default) = runs as the project’s own agent with full tools; non-empty = its own scoped trigger-<slug>-<name> agent with exactly those tools.
run.model	string	Optional per-trigger model override.
run.permissionMode	default | acceptEdits | bypassPermissions | plan	Permission mode the fired turns run under. bypassPermissions runs every tool unprompted — see the caution below.
run.maxTurns	integer	Upper bound on agent turns (default 30).
run.maxSpawnDepth	integer ≥ 0	Bounds internal spawning (0 = may not spawn).
enabled	boolean	Whether it’s armed. A trigger created through the UI or MCP defaults to disabled.

bypassPermissions removes the last check

bypassPermissions runs every granted tool with no prompt and no gate. A schedule fires unattended, on a timer — there is no human at the keyboard to prompt anyway, so the confirmation that would normally catch a bad Bash or Write never happens, and it won’t happen at 3am either. With run.tools, this is the fired agent’s whole capability exercised without review, on every firing.

Use it only where you accept that blast radius unsupervised — a sandbox or a scratch checkout. Otherwise leave it at default or acceptEdits and keep run.tools narrow. See What your agents can do.

Exactly-one rules

A schedule needs exactly one of cron / interval, and its run needs exactly one of prompt / promptFile. A malformed entry in a hand-edited triggers map is dropped (rather than bricking the project), so a bad edit can’t take the project down.

The schedule editor maps one-to-one onto this schema:

The legacy schedules block

Before triggers were unified, schedules lived in their own top-level schedules map. That form is still honored and armed — a simpler shape when you only need a timer:

# project.yaml — legacy standalone schedules (still supported)
schedules:
  nightly-scan:
    type: cron                 # cron | interval
    cron: "0 3 * * *"
    prompt: Scan for dependency advisories.
    resume_session: false      # note: snake_case here (false = fresh chat each fire)
    enabled: true

Note the differences from the unified run block: resume_session (snake_case) in place of session: new|resume, and promptFile here resolves under .paddock/schedules/ rather than .paddock/triggers/. New work is better expressed as a triggers entry (it’s what the Triggers tab and the MCP tools read and write), but existing schedules blocks keep working.

Self-MCP tools

When the trigger-management MCP is enabled — the self-MCP write layer (PADDOCK_SELF_MCP + PADDOCK_SELF_MCP_WRITE) plus PADDOCK_HOOKS_MCP (or a per-project hooksMcpEnabled) — Claude is given four tools. They manage every trigger type; for a schedule, use type: "schedule".

set_trigger

Create or update a trigger (a partial patch — an enabled-only call just flips the toggle). Parameters (note the snake_case MCP argument names):

Parameter	Type	Notes
name	string	The trigger’s stable key. Required.
type	schedule | event | webhook	Omit on an edit to keep the existing when. webhook is reserved (not yet fireable); afterTurn events are reserved for the curator.
cron	string	For a schedule: a 5-field expression (host-local). Exactly one of cron / interval.
interval	string	For a schedule: a duration (30m, 1h).
prompt	string	Inline instruction. Provide this or prompt_file.
prompt_file	string	A .md file under .paddock/triggers/, read at firing.
session	new | resume	new (default) = fresh chat each firing; resume = one owned session.
tools	string / array	Allow-list (one per line, comma-separated, or a JSON array). Empty = the project’s own agent.
model	string	Model override for the fired agent.
permission_mode	default | acceptEdits | bypassPermissions | plan
max_spawn_depth	number	0 = may not spawn.
max_turns	number	Caps the fired turn.
enabled	boolean	Defaults false on a new trigger; omitted on an existing one leaves it unchanged.
project	string	Target project slug; omit for the current project.

list_triggers

List a project’s triggers (all types) — what’s declared, and their state.

remove_trigger

Delete a trigger by name.

run_trigger

Fire a trigger immediately, without waiting for its cron or event — the same “Run now” path the Triggers tab uses. Handy for testing a prompt before you leave it to a schedule. Like the REST route, it starts a real turn, so it is gated with the rest of this family.

REST endpoints

The Triggers tab drives the unified trigger surface. These endpoints are always available (they don’t require the schedule-mutation gate):

Method & path	Purpose
GET /api/projects/:slug/triggers	List triggers + the picker catalog (grantableTools, events, triggerTypes).
GET /api/projects/:slug/triggers/:name	Get one trigger.
PUT /api/projects/:slug/triggers/:name	Create or replace one trigger (full { trigger, run, enabled } record). Enable/disable is this call with enabled flipped.
DELETE /api/projects/:slug/triggers/:name	Delete and disarm one trigger.
GET /api/projects/:slug/triggers/runtime	Just the armed / next-fire runtime state, so the Triggers tab can poll cheaply without re-fetching the config and picker catalog. A static segment, matched before /:name.
POST /api/projects/:slug/triggers/:name/run	Run now — fires through the same hub path a cron or event fire uses. 202 with the session id; 404 for an unknown trigger, 502 if the fire started no chat.

There are no separate schedule endpoints

This page previously listed a parallel set of legacy routes — GET/PUT/DELETE on /api/projects/:slug/schedules[/:name], plus …/:name/:action(enable|disable) and …/:name/trigger — described as still present and governed by the schedule-mutation gate. None of them are registered. The hooks and schedules REST surfaces were retired and collapsed onto the unified trigger routes above, which are the only ones served.

The legacy schedules: block in project.yaml is a different thing, and it does still load — it’s only the HTTP surface that’s gone.

Next steps

• Schedules — the concept.
• Scheduling recurring work — the hands-on guide.
• Scheduling configuration — the deployment gates.