Skills

Skills are Allen Library playbooks that guide how Allen handles a type of request. They help Allen decide whether to answer directly, inspect data, use an agent, or run a workflow—and which safeguards to apply before work starts.

You normally do not need to choose a skill. Allen reviews the enabled skill catalog for non-trivial requests and loads the most relevant playbook. Use an explicit skill when you want a particular operating approach to guide the rest of a conversation.

Media placeholders

Before publishing, capture Settings → Skills, the skill cards and filters, the skill editor, a /skill suggestion in chat, and the compact Skill load message after the command is sent. Suggested files: static/img/docs/concepts-skills-01.png through static/img/docs/concepts-skills-05.png, plus static/videos/docs/concepts-skills-short.mp4 for the chat flow.

What skills change

A skill can guide Allen to:

• gather repository or execution evidence before making a claim;
• choose between a direct answer, data query, specialist agent, or workflow;
• ask for missing scope or confirmation before state-changing work;
• apply task-specific quality, safety, or handoff expectations;
• identify related agents and workflows to verify before routing.

A skill does not perform the work itself. Agents carry out specialist tasks, while workflows coordinate repeatable multi-step processes. The skill tells Allen how to choose and operate those capabilities.

Automatic and explicit skill selection

Method	When to use it	What happens
Automatic selection	Most requests	Allen reviews enabled skill metadata, chooses by intent, and loads the selected playbook without adding routing noise to the conversation.
/skill <name>	You want a specific playbook for this conversation	Allen loads that enabled skill and keeps its guidance in the conversation history. The most recently loaded skill takes precedence where loaded skills overlap.

Search ranking is a discovery aid, not the final routing decision. Allen also considers exclusions, allowed route types, the requested outcome, and the agents, workflows, and tools currently available.

Load a skill in chat

1. Open a new or existing chat.
2. Type /skill to show enabled Allen Library skills.
3. Continue typing to filter by name, then choose the playbook you want.
4. Send the command. The conversation shows a compact Skill load card and Allen acknowledges the playbook.
5. Send the request you want handled under that guidance.

The skill remains available to later turns because Allen preserves the load instruction in the conversation history. Start a new chat when you want a separate outcome without the previously loaded playbook.

If the name is unknown, deleted, or disabled, Allen rejects the load before adding it to the conversation and may suggest nearby skill names. Open Settings → Skills to confirm the slug and enabled state.

tip

Loading a skill does not bypass normal confirmation, permission, or evidence checks. It changes the operating guidance, not the authority Allen has to act.

Browse and manage skills

Open Settings → Skills. The page shows the number of playbooks and enabled skills, plus cards with each skill's display name, slug, description, category, priority, trigger count, and allowed routes.

From this page you can:

• search names, descriptions, categories, and triggers;
• filter the catalog by category;
• enable or disable a skill from its card;
• open a skill to review or edit its full playbook;
• create, duplicate, or delete a skill;
• refresh the list after another operator changes the catalog.

Disable a skill when you want to keep its definition but stop it from normal matching and /skill suggestions. Delete only when the playbook should leave the active catalog. Deleted skills are retained as soft-deleted records; creating the same slug again restores the record instead of creating a duplicate.

Read the skill editor

Area	What it controls
Identity	Stable slug, display name, description, category, and priority. Names are normalized to lowercase hyphenated slugs.
Matching	Trigger phrases that indicate a match, exclusions that argue against it, and related workflow or agent hints.
Routes	Whether the playbook may lead to a direct answer, data query, agent spawn, or workflow run.
Playbook	Markdown guidance Allen loads after selecting the skill. Use Edit, Split, and Preview to review it.
Status	Whether the skill participates in normal selection and can be loaded explicitly.

Priority helps order otherwise similar candidates; it should not be used to force an unrelated playbook to win. Make triggers specific enough to recognize real requests, add exclusions for nearby requests that need a different route, and keep related agent and workflow names current.

Create or revise a skill safely

1. In Settings → Skills, select new skill or open an existing card.
2. Write a clear description of the requests the playbook owns.
3. Add representative triggers and explicit exclusions.
4. Choose only the route types the playbook may recommend.
5. Add related agents or workflows as routing hints, not assumptions that those capabilities still exist.
6. Structure the playbook around when to use it, when not to use it, evidence, clarification and confirmation, capability discovery, routing, and expected output.
7. Preview the Markdown, check the enabled state, and save.
8. Test the result in a disposable chat with both a matching request and a nearby request that should not match.

Saving requires a non-empty normalized name and playbook body. Unsupported route values are removed; when no supported route remains, the record falls back to direct_answer. Updates preserve the skill identity and increment its version.

Good playbook design

• Describe observable user intent rather than a single keyword.
• Separate similar jobs with exclusions, such as investigation versus implementation.
• Require repository or execution evidence when claims depend on current state.
• Tell Allen when to ask for confirmation and what must be shown first.
• Treat related agents and workflows as candidates that must be discovered and verified at runtime.
• Keep private prompts, credentials, customer data, and sensitive internal policy out of user-facing documentation.

Troubleshooting

Problem	What to check
A skill does not appear in chat suggestions	Confirm that it is enabled and not deleted in Settings → Skills.
/skill reports that the skill was not found	Use the saved slug, not only the display name; review any suggested names.
Allen chooses an unrelated playbook	Tighten the skill's triggers and exclusions, lower overly broad priority, and test both positive and negative examples.
A route never gets proposed	Confirm that the route is enabled on the skill and that a matching live agent, workflow, or tool is actually available.
Saving fails	Check that the slug and body are present and that another active skill does not already use the same slug.

Related concepts: Agents, Teams, and Workflows.