Skip to content
Esc
↑↓navigate↵open⌘Jpreview
On this page

Agent Skills

Teach the agent how your team works with reusable instructions

Skills are reusable instructions you write once and the Modem Agent can pull in whenever a relevant task comes up. Each skill is a markdown document that captures how your team handles a specific workflow: an escalation playbook, a release notes format, a bug triage checklist, the steps you take when a topic mentions a security report.

Skills are private to your organization. Other Modem customers cannot see them, and the agent only loads them inside your organization’s chats and automations.

How It Works

Each skill has three parts:

Field Purpose
Slug A short, unique name for the skill (e.g. bug-triage-checklist). Lowercase letters, numbers, and hyphens.
Description A one-line summary of when to use the skill. The agent reads this to decide whether the skill is relevant.
Body The full instructions in Markdown. Headings, lists, examples, links to internal docs, whatever helps the agent follow your process.

When you chat with the agent, it sees the slug and description of every skill in your organization. If a skill looks relevant to what you’re asking, the agent reads the full body and follows the instructions. You can also reference a skill by name in your prompt to make sure it gets used.

What Skills Are Good For

Skills work best for workflows your team repeats. A few examples:

  • Triage playbooks. “When a topic mentions a billing issue, look up the customer’s plan, check for related Linear tickets, and post a summary in #support-leads.”
  • Output formats. “Release notes should always include a ‘For Customers’ section with non-technical language and a ‘For Internal’ section with implementation details.”
  • House style. “When summarizing user feedback, never use the word ‘users’: say ‘customers’ or ‘developers’ depending on context.”
  • Source-of-truth links. “Our PR review checklist lives at [internal URL]. Always reference it when asked about code review.”
  • Multi-step processes. “When asked to draft a customer email, write a first draft, check it against our tone guide, and ask me to confirm before sending.”

If you find yourself pasting the same context into the agent every time you ask a similar question, that context belongs in a skill.

Creating a Skill

Skills live in Settings → Skills. Only organization admins can create, edit, or delete skills. Members can see the list and trigger skills through the agent.

There are two ways to create one.

With the agent

The fastest way is to describe what you want and let the agent draft it.

PromptCreate a skill for triaging incoming bug reports. It should check for duplicates in Linear, gather any related topics from Modem, and post a summary in #bug-triage.

The agent will write a draft skill and confirm before saving. You can edit it afterwards from the Skills page.

Manually

Open Skills

Go to Settings → Skills in your dashboard and click Add skill → Create manually.

Pick a slug

Use a short, descriptive name like release-notes-format or customer-escalation. Up to 60 characters.

Write the description

One sentence that tells the agent when this skill applies. Up to 240 characters. The agent uses this to decide whether to load the full body, so be specific.

Write the body

Plain Markdown. Headings, lists, code blocks, and links all work. Up to 20,000 characters. Treat it like a runbook a new teammate would follow.

Save

Click Create Skill. The skill is immediately available to the agent for everyone in your organization.

Writing Effective Skills

A few patterns that work well:

  • Lead with when to use it. Start the body with a sentence or two on the trigger conditions, even though the description already covers it. The agent benefits from the reminder once it has the full text.
  • Reference exact tool and field names. If you want the agent to update a Linear issue’s priority, say “set the priority field” rather than “update the urgency.”
  • Show the desired output shape. If the skill produces a summary, message, or document, include a short example of what it should look like.
  • Note approval and safety rules. If certain steps need human confirmation (“ask me before posting in #announcements”), say so explicitly.
  • Keep it focused. One skill per workflow. If a skill grows past a few thousand characters, consider splitting it into related skills the agent can compose.

Managing Skills

The Skills page has two tabs. Organization lists the skills your organization created. Built-in lists the skills Modem ships, covered below. The page opens on Organization.

Use the search box above the list to filter the open tab. If nothing matches there but the other tab has matches, a link takes you to them.

The Organization tab shows each skill’s slug and description, and under Last updated, who last edited it and when. Click a row to open the skill. Admins can edit its slug, description, and body there; members can read them.

To delete a skill, open it and click Delete skill at the bottom of its page. Modem asks you to confirm, and the agent stops using the skill immediately.

Updates take effect on the next agent message. There is no separate publish step.

Built-in Skills

Modem’s built-in skills appear on the Built-in tab in Settings → Skills, in three groups:

  • Workflows, such as Weekly digest, Reporting digests, Filing tickets, and User health.
  • Integrations, such as GitHub, Linear, Sentry, and Web search.
  • Using Modem, such as Automations, Stories, and Charts.

Each workflow and Using Modem row shows the skill’s name and summary. When an automation template uses the skill, the row links to it: 1 template opens that template, and N templates opens the template gallery filtered to those templates.

Each integration row shows the integration’s status:

  • Connected when your organization has connected it.
  • Always on for Web search, which needs no setup.
  • Connect when you haven’t connected it yet. It opens that integration’s settings.

Automation templates name built-in skills in their prompts by the same name the row shows. For example, a prompt that says “Use the reporting-digests skill.” refers to the reporting-digests row.

You can’t edit or turn off a built-in skill, or view its full instructions. Skills you create sit alongside the built-in set and never override Modem’s safety rules.

Was this page helpful?