Skills
Skills are reusable packages of instructions, templates, and tools that teach Glean how to execute specific tasks with domain expertise. Instead of prompting from scratch every time, use a Skill that captures the structure, decision logic, and know-how behind a task and pairs that knowledge with the tools needed for execution.
General-purpose AI models are powerful but can struggle with domain-specific workflows like account planning, Salesforce analysis, and contract reviews. Skills close this gap by making expertise repeatable and composable, and by loading that expertise only when relevant.
Skills follow the open Agent Skills standard. This means you can bring compatible Skills from any source that follows the standard into Glean and layer in your own organization's ways of working.
Available skills can also be reused in auto mode agents, so you can apply the same task-specific guidance in both Glean Assistant and auto mode agents.
Key concepts
Glean Skills
Glean Skills are built-in Skills that Glean runs under the hood to power Assistant capabilities. You won't see most of them in the Skills library or Shared tab. When Assistant invokes a Glean Skill while answering you, the Skill appears as an intermediate step you can click into to see what it did. Those skills still run when an admin turns Skills off for your organization.
Deep Research is a Glean Skill you start yourself, so it follows Skills access. If you don't have Skills access, Deep Research isn't in the skill picker and /deep-research doesn't run. See Deep Research.
Personal Skills
Create your own Skills tailored to your workflows. Personal Skills are visible only to you and can be enabled, disabled, edited, or deleted at any time from Settings → Skills.
Shared Skills
Skills that teammates share with you, or that your organization publishes for a team or the whole company. Shared Skills appear in a separate Shared area so you can tell them apart from your own. You can use a Skill someone shares with you without recreating it, and you stay on the owner's latest version as they make updates. See Sharing Skills.
Skill availability and enablement
For an Assistant to invoke a Skill for a given query, the following conditions govern access, control, and state management:
- Recipient enablement required: A recipient must explicitly keep the Skill enabled in their settings for the Assistant to invoke it during their sessions.
- Independent creator controls: Disabling a Skill as a creator affects only the creator's account. It remains fully accessible and active for recipients who have valid access and keep it enabled. To use the Skill again, the creator must re-enable it.
- Enforced access and safety overrides: Recipient enablement cannot bypass access permissions or security restrictions. Draft Skills and security-blocked Skills remain permanently unavailable regardless of recipient toggle states.
Automatic Skill routing
Glean automatically evaluates your query against available Skill descriptions and determines whether a Skill is relevant. If it is, Glean loads the relevant portions of the Skill to inform its response. You can also explicitly reference a Skill by name in your prompt.
On the Glean Universal Model Key, Skill routing works in Auto and with any model you pick. On a Customer Key, Skill routing requires Thinking mode. Skills aren't available in Fast mode, so neither automatic routing nor referencing a Skill by name works there. See Skills aren't available in Fast mode.
Interoperability
Skills that follow the Agent Skills standard can be imported directly into Glean. You can also import Skills from public and private GitHub repositories using a URL.
Getting started
Accessing Skills
Go to Settings → Skills in your Glean instance. You'll see two tabs:
- Personal — Skills you've created.
- Shared — Skills teammates or your organization have shared with you.
Your admin decides who can create Skills. To create one in chat, upload a file, or import from GitHub, you need creation permission in addition to Skills access. Without that permission, those options are hidden. You can still use Skills shared with you or published to an audience that includes you.
Uploading a Skill
- Click Add in the top-right corner of the Skills page.
- Choose Upload a Skill and select a
.zip,.md, or.skillfile. - The Skill appears under the Personal tab once uploaded.
If you upload a Skill with the same name as an existing one, Glean prompts you to replace the existing Skill instead of creating a duplicate.
Importing a Skill from GitHub
You can import a Skill directly from a public GitHub repository or a private repository that your GitHub account can access. Glean fetches the Skill, validates it, and records the source so the Skill can stay in sync with upstream changes.
To preview, import, or sync a GitHub Skill, you need Skills access and permission to create Skills. Your admin must also turn on GitHub imports. If Skills is limited to a test group, teammates outside that group can't use GitHub Skills. See Set up Skills and manage access.
- Go to Settings → Skills and click + Add Skill.
- Select Import from GitHub.
- Paste a GitHub URL. Supported formats include:
- A repository root URL (for example,
https://github.com/org/repo) - A subtree path (for example,
https://github.com/org/repo/tree/main/skills/my-skill) - A direct link to a SKILL.md file (for example,
https://github.com/org/repo/blob/main/SKILL.md)
- A repository root URL (for example,
- Click Preview. Glean fetches the Skill and shows the parsed name, description, file tree, and SKILL.md content.
- Review the preview and click Import.
- The Skill appears under your Skills with an Imported from GitHub badge.
Glean imports the SKILL.md file along with any bundled supporting files such as templates, reference materials, and scripts. A single Skill can contain up to 100 files across up to 6 directory levels.
Use a branch or tag URL when importing. Commit permalink URLs that point to a specific commit aren't supported because they can't stay in sync with upstream changes.
Private repository imports require GitHub authorization. If prompted, connect your GitHub account and grant Glean access. Glean can import only repositories and paths that your GitHub account can read.
Import multiple skills from a directory
If a GitHub URL points to a directory that contains more than one Skill, Glean recursively discovers every Skill under that path and imports them together. This works at any nesting depth, so you can point to a repository root or any parent folder and Glean finds every SKILL.md below it.
Glean treats each folder that has its own SKILL.md as a separate Skill, regardless of how many levels deep it sits. For example, a repository might organize Skills across multiple directories:
repo/
├── skills/
│ ├── account-planning/
│ │ └── SKILL.md
│ └── meeting-prep/
│ ├── SKILL.md
│ └── templates/
└── plugins/
└── diligence/
└── skills/
└── risk-analysis/
├── SKILL.md
└── scripts/
Pointing Glean to the repository root discovers all three Skills, including risk-analysis nested several levels deep.
To import them:
- Go to Settings → Skills and click + Add Skill.
- Select Import from GitHub.
- Paste the URL of the repository or directory that contains the Skills (for example,
https://github.com/org/repoorhttps://github.com/org/repo/tree/main/skills). - Click Preview. Glean lists each Skill it finds under that path, along with its parsed name and description.
- Review the discovered Skills and click Import.
- Each Skill appears under your Skills as a separate entry, with its own Imported from GitHub badge.
If a Skill in the directory can't be imported, Glean skips it and imports the others, so one invalid Skill doesn't block the rest. A Skill is skipped if its SKILL.md is missing the required name or description frontmatter, or if the Skill exceeds the size limit.
Multi-Skill import works the same way as single-Skill import, with a few differences:
- A URL that points to one Skill, such as a Skill folder or a direct SKILL.md link, imports a single Skill.
- A URL that points to a directory of Skill folders imports all of them at once.
- Skills you've already imported from the same source URL aren't imported again. Sync those Skills instead to pull in upstream changes.
Each imported Skill is independent. It keeps its own source link and sync status, and you can enable, edit, share, or delete it separately.
Keeping imported Skills up to date
When you import a Skill from GitHub, Glean records the source URL and the commit it was imported from. Glean automatically checks the source repository for upstream changes about once a day and pulls any updates into your imported Skill.
Each imported Skill shows one of the following sync statuses:
- Up to date: The Skill matches the latest content in the source repository as of the most recent sync.
- Sync failed: Glean couldn't reach the source. This can happen if the repository or file was moved, renamed, deleted, or your GitHub authorization no longer has access to it.
Sync is one-way: changes flow from GitHub into Glean, not the other way around. If you edit an imported Skill in Glean, those changes stay in Glean and aren't pushed back to the source repository.
Creating a Skill via chat
If you have permission to create Skills, you can create them by chatting with Glean:
- Describe the workflow you want to turn into a Skill. For example, "Create a Skill for writing weekly status updates."
- Glean asks clarifying questions about the task, then scaffolds the Skill in canvas.
- Review the generated Skill, make any edits, and save it. The Skill appears under the Personal tab in Settings → Skills.
Using a Skill in chat
There are multiple ways to use Skills:
- Automatic: Ask Glean a question related to a Skill's domain. If a relevant Skill exists, Glean automatically loads and uses it. On a Customer Key, use Thinking mode, as Skills aren't available in Fast mode.
- Explicit: Reference a Skill by name in your prompt (for example, "Use my account planning Skill to build a plan for Acme Corp").
- Slash command: Type
/in the composer to see your available Skills and invoke one directly (for example,/my-skill-name). - From the composer: Click
+in the composer and select a Skill to view it or invoke it directly.
Managing your Skills
From Settings → Skills, use the toggle and the three-dot menu on each Skill card to:
- Enable/Disable — Toggle Skills on or off.
- Edit — Update Skill instructions at any time.
- Share — Share a Skill with teammates, departments, or groups. See Sharing Skills.
- Replace — Re-upload a Skill to update its contents.
- Delete — Delete a Skill you own. Editors of shared Skills can't delete them. Admins and Skills Moderators can delete shared Skills within their scope from the Admin console. If Delete isn't available, you may not have permission. See Delete a shared Skill.
- Download — Export any Skill as a
.zipfile for portability.
For a shared Skill, the toggle controls only your own ability to invoke it. Turning it off does not remove it from other recipients or change the creator's or other recipients' settings.
Admin controls
Admins set Skills access, creation, and usage from Admin console → Skills → Setup. They can also turn GitHub imports on or off.
- Whether Skills are turned on, and for whom.
- Who can create Skills.
- Who can use Skills.
- Whether you can share Skills, and who you can share them with.
- Which Skills are published to a team or the whole organization.
Admins can pilot Skills with a test group. See Roll out Skills to a test group.
For roles and sharing, see Set up Skills and manage access and Share Skills.
Deep Research
Deep Research is a Glean Skill that writes a cited report from your company's sources and the web. Use it to ensure a chat answer keeps the comparisons, timelines, or sources behind a conclusion. A report usually takes 5–30 minutes to complete and outputs 5–10 pages, including linked citations.
Choose chat or Deep Research
Chat is the better fit for a single fact, a short rewrite, or any task that has to change a connected app. Deep Research only reads. It won't file a ticket, post a message, or update a record.
These questions are a better fit for a report than for chat:
| Goal | What to ask for |
|---|---|
| Track a project | Recent issues for one customer, with status, pulled from Jira, Gong, and Slack |
| Compare a competitor | New launches in one feature area, from their blogs, marketing pages, and news |
| Plan a bet | Untapped demand in one business area, from Gong calls, Slack threads, and Salesforce activity |
Prerequisites
Complete the following prerequisites:
- Have Deep Research turned on for your account. See Deep Research.
- Have Skills access. See Set up Skills and manage access.
Turning Deep Research on doesn't grant Skills access. If you don't have it, Deep Research isn't in the skill picker and /deep-research doesn't start a report. Ask your admin to turn Skills on for you.
On the Glean Universal Model Key, the separate Deep Research monthly quota doesn't apply. Usage follows your standard Assistant terms.
Start a report
- Open Chat.
- In the composer, open the skill picker and select Deep Research. You can also type
/deep-researchand the question. - State the goal. Name the teams, systems, or documents that matter.
- Submit. The report runs in the background, and Glean notifies you when it's ready.
A question can look like this:
/deep-research Summarize competitor launches in Q3 across their blogs, press releases, and our Gong calls.
Run Deep Research on a Customer Key
If Deep research is still in the mode picker, Deep Research is still a chat mode for your organization. Select Deep research, then enter the question. The chat mode doesn't need Skills access, and the monthly quota still applies.
Where Deep Research runs as a skill on a Customer Key, start a report from the skill picker or with /deep-research. Skills access is required.
Pick up an older chat
A thread that ran on the previous Deep Research agent shows this banner: "Deep Research is now a skill. Start a new chat and add /deep-research to your question to run it." Start new chat places /deep-research in the composer. The new chat still needs Skills access.
Choosing Try again or Think deeper starts a normal chat and drops /deep-research. Type the command again if you still want the report.
See what a report includes
Glean searches your connected systems and the web, then writes one report with linked citations. Web pages come from Brave Web Search. You only see documents you already have permission to open.
Deep Research supports the Glean Universal Model Key and Customer Key (BYOK). It uses the model selected for that chat. On the Glean Universal Model Key, that's the model Glean routes to. The skill is available in the web app.
Sharing Skills
Once you've built a Skill that works well, you can share it with teammates so they benefit from the same expertise without sending files back and forth or maintaining stale copies.
Your administrator controls whether you can share Skills and who you can share them with. If sharing options are unavailable, your admin hasn't enabled sharing for your account.
Share a Skill
Open the share dialog
Go to Settings → Skills, find the Skill you want to share, click its three-dot menu, and select Share.
Choose who to share with
Under People with access, add individual teammates, departments, or groups. You can also use Copy link to send a direct link to anyone who has access.
Set a permission level
Choose a permission level for each recipient:
- Viewer — Can use the Skill.
- Editor — Can use and edit the Skill.
You are the Skill's Owner by default. Owner can't be assigned from the share dialog.
Optionally allow company-wide access
Under General access, switch from Restricted to people with access to Anyone at [Company] with the link can view to let anyone in the organization find and use the Skill. This option requires admin permission.
Recipients are notified by email when you share a Skill with them, and again when you update a Skill they've added — so they always have your latest version. Admins can publish Skills to the whole organization from the Admin console.
Use Skills shared with you
Skills that teammates share with you appear in the Shared tab of Settings → Skills, separate from the Skills you've created.
- Add a shared Skill to start using it. Adding a Skill keeps it available in chat and keeps you on the owner's latest version.
- Keep the Skill turned on to let Assistant invoke it for your queries. You can turn it off at any time without affecting the creator or other recipients.
- Skills your organization publishes may already be enabled for you, depending on how your admin set them up.
- You'll receive an email when a Skill you've added is updated, so you can review what changed.
When Skill names overlap
If a Skill shared with you has the same name as one of your Personal Skills, your Personal Skill takes precedence. Glean keeps both available and adjusts the name of the shared Skill so you can tell them apart.
Skill file format
A Skill is structured as a folder containing:
- SKILL.md (required): The main instructions file with YAML frontmatter (
nameanddescriptionfields) and natural language instructions. - Additional files (optional): Templates, examples, scripts, and reference materials.
Example use cases
| Use case | Description |
|---|---|
| Account planning | Upload a Skill that encodes your organization's account planning best practices. Glean uses it to generate standardized, high-quality account plans grounded in Salesforce, Gong, and internal docs. |
| Meeting prep and follow-up | Create a Skill that pulls from your calendar, email, Slack, and CRM to generate meeting prep briefs or draft follow-up emails with the right tone and context. |
| Domain-specific analysis | Teach Glean your team's methodology for financial analysis, legal review, or technical documentation so outputs follow your standards consistently. |
| Knowledge base article creation | Use a Skill to standardize how support articles are written, ensuring consistent formatting, tone, and completeness from resolved tickets. |
| Daily briefings | Build a personal Skill that synthesizes your calendar, email, Slack, and task list into a morning brief you can reference throughout the day. |
Skills vs. Agents
Skills and Agents are complementary features. Here's how to think about when to use each:
| Skills | Agents | |
|---|---|---|
| What they are | Reusable packages of instructions and expertise for specific tasks | End-to-end workflows triggered by events or schedules |
| When to use | When you want consistent, reusable expertise applied to a task on demand | When you want to automate a full process that runs autonomously |
| Example | A market research Skill that applies consistent methodology every time | An agent that runs monthly, builds a market report using a Skill, and emails it to leadership |
Skills are foundational building blocks that Agents (and Glean) can invoke. They work together — you don't have to choose one over the other.
Permissions and data access
Skills operate within Glean's existing permission model. Glean will only access data that you have permission to view. No new data access is granted by enabling Skills.
Skill safety and scanning
Glean scans Skills you upload locally and Skills imported from a third-party repository for risky content, including unsafe instructions and malicious code. Skills from Glean are safe by default and aren't scanned. A scan runs when you upload or import one of those Skills, and again whenever its instructions or files change. Scanning happens in the background, and your admin can turn it off for your organization.
A Skill stays usable while its scan runs, and you may see a short notice that a scan is in progress. A Skill becomes unusable only when a moderator turns it off.
Warnings on a Skill you own
If a scan reports findings on a Skill you own, you see a Review recommended warning that lists each finding. You can:
- Select Fix with Glean to apply a suggested fix. This saves a new version and starts a new scan.
- Select Edit, change the Skill, and save. Saving starts a new scan of the updated content.
- Select Dismiss findings to hide the warning for yourself. Dismissal applies only to you and only to the current version. It doesn't hide the warning from others, and it returns if the content changes.
Findings clear once a new scan completes. A moderator can review your Skill and turn it off if needed.
Skills turned off for security
A moderator can turn off a Skill they consider unsafe. If a Skill you have access to is turned off, you see a short notice that it's temporarily turned off for security reasons. A turned-off Skill is unavailable to everyone until its owner saves a new version or a moderator turns it back on. Saving a new version starts a new scan and restores use unless a moderator turns off that version too.
For how moderators review findings and turn off Skills, see Skill scanning and safety review.
Tips for writing effective Skills
- If a Skill isn't triggering when you expect it to, Skill routing depends on how well the Skill's name and description match the user's query. Try refining the Skill's description to include clearer keywords and, optionally, negative examples (cases where the Skill should not be used).
- Skills work best with advanced reasoning models. On the Glean Universal Model Key, use Auto or pick a model from For complex tasks. On a Customer Key, use Thinking mode for the best Skill routing and execution quality.
Reflect on a Skill
Skill reflection turns feedback you give a Skill in chat into a proposed update to that Skill. After about 10 learnings, Glean asks clarifying questions and suggests an update, or you can type /reflect yourself. You review the feedback behind each proposal and approve it before it applies. See Skill reflection.
Current limitations
- Skills aren't available in Fast mode: In Fast mode, Glean doesn't load Skills at all. Both automatic routing and explicit references to a Skill by name have no effect, and Glean may report that the Skill isn't available to it. This is not a permissions issue. Use thinking mode when you need a Skill to run. In Adaptive mode, a Skill runs only when Glean escalates the request off the fast path, so Skill use is not guaranteed.
- GitHub access follows source permissions: Importing or refreshing Skills from private GitHub repositories requires an active GitHub authorization with access to the source repository.
- GitHub sync is one-way: Changes flow from GitHub into Glean, not from Glean back to GitHub. If you edit an imported Skill in Glean, those changes stay in Glean.