Skip to content

rundesk-ai/rundesk-cli / src/skills/managing-rundesk

managing-rundesk

Verified

Use 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.

references/skills.md

View on GitHub
# Skills

Skills are filesystem instruction bundles in Rundesk's library. Catalogs supply skills; grants make
selected skills visible to an agent; profiles configure the values a skill declares.

## Library and catalog commands

| Command | Control |
|---|---|
| `skills` or `skills list [<agent>]` | List the library or one agent's grants. |
| `skills catalogs` | List installed catalogs, versions, and sources. |
| `skills install <repository> [--confirm]` | Validate and preview a local directory or GitHub catalog; `--confirm` installs it. |
| `skills update <catalog> [--confirm]` | Compare with its source; `--confirm` replaces catalog content when changed. |
| `skills remove <catalog> [--confirm]` | Preview removal; `--confirm` removes the catalog and its grants. |

## Grant and value commands

| Command | Control |
|---|---|
| `skills grant <agent> <catalog/skill> [--as <name>]` | Present a skill to an agent, optionally under an alias. |
| `skills revoke <agent> <skill>` | Remove one grant as it stands in that agent. |
| `skills profiles <catalog/skill>` | List configured accounts without revealing values. |
| `skills configure <catalog/skill> [--profile <name>]` | Prompt at the terminal for the values that account needs. |
| `skills forget <catalog/skill> [--profile <name>] [--confirm]` | Preview emptying one account; `--confirm` forgets it. |
| `skills doctor [<agent>]` | Diagnose all grants or one agent's and print exact fixes. |

Prefix every command with `"$RUNDESK_COMMAND"`.

## Optional first-party catalogs

Check `skills catalogs` first. If the requested capability is absent, preview the relevant catalog:

- Apple Calendar, Contacts, Mail, and Messages on macOS:
  `https://github.com/rundesk-ai/rundesk-skills-apple`
- Guarded service integrations:
  `https://github.com/rundesk-ai/rundesk-skills-integrations`
- Rundesk tasks, inbox, mentions, projects, pages, and assets through the `desk` API client:
  `https://github.com/rundesk-ai/desk-cli`

```sh
"$RUNDESK_COMMAND" skills install https://github.com/rundesk-ai/rundesk-skills-apple
"$RUNDESK_COMMAND" skills install https://github.com/rundesk-ai/rundesk-skills-apple --confirm
"$RUNDESK_COMMAND" skills install https://github.com/rundesk-ai/rundesk-skills-integrations
"$RUNDESK_COMMAND" skills install https://github.com/rundesk-ai/rundesk-skills-integrations --confirm
"$RUNDESK_COMMAND" skills install https://github.com/rundesk-ai/desk-cli
"$RUNDESK_COMMAND" skills install https://github.com/rundesk-ai/desk-cli --confirm
```

Treat the unconfirmed preview as authoritative for the catalog's current skills. Add `--confirm`
only after its source, name, and skills match the request.

An agent turn may run the confirmed command when the owner authorized that effect and its configured
tool access can invoke Rundesk. The turn environment grants no authority by itself; the ordinary
preview, confirmation, validation, and locking guards still apply.

The `desk-cli` catalog installs the agent instructions; it does not install the `desk` binary.
Follow that repository's README to install or update the executable separately, and verify
`desk --version` before granting its skill:

```sh
"$RUNDESK_COMMAND" skills grant <agent> desk-cli/managing-your-desk
"$RUNDESK_COMMAND" skills profiles desk-cli/managing-your-desk
"$RUNDESK_COMMAND" skills configure desk-cli/managing-your-desk --profile <name>
"$RUNDESK_COMMAND" skills doctor <agent>
```

Configure one complete named profile for each API identity the agent may use. A desk-bound identity
uses `desk inbox` and `desk mentions` for exactly its desk. A non-desk identity acts as the signed-in
human and uses `desk user-mentions` for that human inbox; those are separate from mentions addressed
to the API-token actor. Check `desk account --json` rather than inferring authority from a profile
name: Owner and Admin may target and manage desks, while a Member is limited to its assigned visible
desk; a deskless Member retains human mentions but has no task or project desk scope. Prefer the
least-privileged identity that can do the requested work, and remember that every agent holding the
skill can reach every profile configured for it.

## Safe workflow

1. Run `skills catalogs` and `skills list` before changing the library.
2. Preview catalog install, update, or removal without `--confirm`; repeat with it only after source,
   catalog name, changed files, and affected grants match the request.
3. Grant by full `catalog/skill` address. Use `--as` only to resolve a name collision or present two
   copies under distinct names.
4. Configure values at the owner's terminal, never through chat. A named profile is a separate
   account; no `--profile` means the default account.
5. Run `skills doctor <agent>` after each grant, profile change, catalog update, or credential fix.
   A value set during the current turn reaches the next turn, not the already-running process.

## Catalog and grant behavior

- Install and update may read a local directory or reach GitHub. The catalog source remains the
  source of truth: update replaces installed catalog content, including local edits to that fetched
  copy.
- Removing a catalog also revokes grants pointing into it. Save a backup and inspect the preview.
- The shipped `rundesk` catalog and required `skills` catalog cannot be removed. Rundesk
  never replaces or removes the owner's `local` catalog. `rundesk-skills`, required before 0.62,
  is an ordinary catalog that `skills remove` may take.
- `managing-github` is granted from `skills/managing-github`; the shipped catalog no longer holds it.
- Revoking a skill removes the agent's presentation, not the library copy or stored profiles.
- Rundesk manages the bundled `delegating-work` grant with outbound delegation scope: new agents and
  agents configured for an exact or unrestricted scope hold it; `--delegate-to-none` removes the
  bundled grant but leaves an owner-managed entry of the same name alone. Do not manually grant it
  back to an inbound-only specialist.
- Every agent granted a skill can reach every configured profile for it. Profiles choose coherent
  account value sets; they are not per-agent access control and never fall back to default values.
- `forget` empties values for only the selected skill account. It requires preview and `--confirm`.
- `doctor` checks granted skills. A library skill that is not granted has no agent context to
  diagnose.

## Create or publish skills

The commands above manage skills that already exist. Use
[Writing skills](../../writing-skills/SKILL.md) to create, revise, review, debug, or publish one; it
routes to its integration and publishing references only when they are needed.

Rundesk has no `skills create`, `skills validate`, or `skills publish` commands.
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: 472kUpdated Claude CodeCodexVerifiedInstall
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: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
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 availableStars: 322kUpdated Claude CodeCodexVerifiedInstall