rundesk-ai/rundesk-cli / src/skills/managing-rundesk
managing-rundesk
VerifiedUse when managing this agent, any other agent in its Rundesk install, or the install itself. It supplies verified operational workflows and safety rules for agent instructions, gateways, conversations, channels, schedules, providers, skills, backups, configuration, credentials, and the Rundesk lifecycle. Do not use for ordinary project work or for deciding, performing, or reviewing delegation.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 5.0 KB |
| references/agent-instructions.md | markdown | 12.3 KB |
| references/agents.md | markdown | 5.9 KB |
| references/backups.md | markdown | 2.1 KB |
| references/channels.md | markdown | 6.6 KB |
| references/configuration.md | markdown | 3.6 KB |
| references/conversations.md | markdown | 5.3 KB |
| references/gateways.md | markdown | 3.1 KB |
| references/lifecycle.md | markdown | 3.7 KB |
| references/maintenance.md | markdown | 5.9 KB |
| references/oauth-login.md | markdown | 3.4 KB |
| references/permissions.md | markdown | 3.6 KB |
| references/providers.md | markdown | 2.8 KB |
| references/retrospective.md | markdown | 3.8 KB |
| references/schedules.md | markdown | 6.4 KB |
| references/self-improvement.md | markdown | 10.7 KB |
| references/skills.md | markdown | 6.6 KB |
| references/teams.md | markdown | 3.2 KB |
references/schedules.md
View on GitHub# Schedules
A schedule belongs to one agent and starts either a program or an agent prompt when local time is
due. The gateway is the clock that fires it.
## Commands
| Command | Control |
|---|---|
| `schedules` or `schedules list [<agent>]` | List schedules that can still run, for every agent or one. |
| `schedules list [<agent>] --expired` | List only schedules that can never run again. |
| `schedules add <agent> <schedule> (--when <cron> | --at <moment>) (--run <program> | --ask <prompt>) [--until <moment>] [--disabled] [--channel <channel> --to <id>]` | Create repeating or one-time work. |
| `schedules update <agent> <schedule> [--when <cron> | --at <moment>] [--run <program> | --ask <prompt>] [--until <moment>] [--enable | --disable] [--channel <channel> --to <id>]` | Change stated fields and keep execution history. |
| `schedules show <agent> <schedule>` | Show the complete definition and state. |
| `schedules run <agent> <schedule> [--wait <seconds>]` | Run now in this terminal; default wait is 3600 seconds. |
| `schedules remove <agent> <schedule>` | Remove the schedule. This has no confirmation preview. |
Prefix every command with `"$RUNDESK_COMMAND"`.
`weekly-self-improve-upkeep` is Rundesk's protected per-agent policy, not an owner cron. It becomes
due after seven distinct usage dates and is controlled only with `agents configure <agent>
--self-improve <true|false>`. Inspect it with `schedules list` or `schedules show`; do not try to add,
update, run, disable, or remove it. A pre-policy owner schedule already using that name stays
owner-controlled and blocks automatic upkeep until the owner removes it; Rundesk never adopts it.
## Use schedules to close future verification loops
Use a schedule proactively when an outcome can only be confirmed later: check that a release stayed
healthy, a gateway came back after an update, an external job finished, or a temporary fix still
works. Prefer one `--at` check for one open loop; use repeating `--when` only for an ongoing need.
Verify an outcome now when it is already observable instead of scheduling avoidable delay.
Make an `--ask` prompt executable on fresh context: name what to inspect, the expected result, the
evidence to use, and who should hear about success or failure. Ask the agent to perform the check,
not merely remind the owner that a check was planned. Keep the check within the original authority;
a schedule changes when work runs, not what changes it is allowed to make. Configure a notified
channel when the result must reach the owner proactively.
## Send one schedule's report somewhere of its own
Name `--channel` and `--to` together to send that schedule's notice and report to one place or one
person instead of the agent's notified channel. The agent's notified channel does not move. Omit
both to keep it.
- Write `--to` the way an allow-list entry is written: a bare sender id for that person's direct
message, `place:<id>` for that place. The destination must already be on that channel's allow
list, as the same kind of thing; check with `channels show <agent> <channel>` first, because a
destination held as the other kind is refused rather than reinterpreted.
- One flag without the other is refused, and so is a channel the agent does not have or an adapter
that does not say it can address a destination. Every refusal happens before anything is written.
- A place target puts the report in a thread on its own notice; a direct message does not. Neither
posts anything between the two messages.
## Time and execution rules
- `--when` is five-field cron: `minute hour day month weekday`. `--at` is
`YYYY-MM-DDTHH:MM`. Both use the machine's local clock; timezone suffixes and `Z` are refused.
- Use exactly one of `--when` or `--at`, and exactly one of `--run` or `--ask`.
- `--until` is the local moment after which the schedule never runs again.
- `--run` takes one quoted string, splits it into program arguments, and never invokes a shell. It
does not expand variables, globs, pipes, redirects, `&&`, or `;`. Name an executable program.
- `--ask` uses a schedule-specific conversation, separate from the conversation used by `ask`.
## Arrange the next trigger
Ending a turn ends the work; a task brief or memory preserves context but does not wake the agent.
When an owner authorizes an ongoing outcome, or the agent commits to finish something later today or
on a future day, arrange the next trigger before the turn ends. Use `--at` for one promised time
block and `--when` for a genuinely recurring responsibility. The scheduled prompt must carry the
original authority, exact next action, limits, and proof. State the exact local start or delivery
time to the owner; do not substitute an open-ended window such as “later” or “today.” The owner
should not have to return merely to restart work already promised.
Do not create a schedule for an aspiration, an unapproved state change, or work with a missing
decision. A schedule preserves timing, not authority. Verify the saved definition. An enabled
self-schedule is accepted only from a turn whose gateway is currently known running; `add` and
`update` refuse it before writing otherwise. For an agent resumption use `--ask`; `--run` starts a
program instead. A successful enabled `--ask` self-schedule establishes the resumption path without
another inspection. Do not report routine gateway state or a possible later outage as a current
blocker. After `show` verifies that schedule, stop inspecting. Unless the owner asks for mechanics,
the final states the verified current result and what the scheduled work will produce and when. Add
a current material blocker only when it affects that continuation. Do not name commands, gateways,
or other routine continuation machinery unless the owner asks.
## Operate and recover
1. Add disabled when rollout needs inspection, then `show` and `run` it manually.
2. Manual `run` takes the same durable claim as the gateway and refuses a duplicate. It records the
result but does not consume or shift the next due time.
3. Enable only after the manual result is correct.
4. Inspect `show`, `gateways logs <agent>`, and `turns <agent>` for failures.
Rundesk durably claims work before starting it and does not replay every minute missed while a
gateway was down. Disable a schedule to retain its definition without future firings. Because
`remove` acts immediately, verify agent and schedule names first and save a backup when the history
matters.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |