Skip to main content
Skills are reusable packages of instructions and optional supporting files. They let your team define a repeatable way to complete a task once, then apply it consistently across Flowglad. Each Skill has a name, a slash-command slug, a description, and instructions written in Markdown. Supporting files can provide additional context or resources needed to follow those instructions.

​
Use Flowglad as your team’s Skill library

Flowglad keeps each Skill as a versioned package. Publishing a change creates a new revision instead of replacing the previous package in place. Organization admins can open a Skill’s History to inspect its revisions, compare changes, restore an earlier revision, or discard a draft when those actions are available. This gives your team one governed place to maintain reusable instructions while preserving which version was active at a given time. Draft and rejected revisions do not replace the active version.

​
Manage Skills through MCP

After you connect the Flowglad MCP server, an agent can inspect and manage your organization’s Skill library for you:
  • skills.list returns the Skills visible to the connection.
  • skills.get returns the latest and active revision metadata for one Skill.
  • skills.create submits a new Skill package.
  • skills.create_candidate submits a new revision of an existing Skill.
  • skills.publish_revision verifies and publishes a deferred revision.
  • skills.create_package_upload_session and skills.finalize_package_upload_session transport packages that are too large to send inline.
  • skills.get_lifecycle_request reports the accepted request’s verification and promotion result.
Creating or updating Skills requires the skills:write permission on both the teammate’s organization membership and the MCP connection’s grant. If you add the permission after connecting the server, reconnect Flowglad MCP and authorize skills:write so the write tools become available. Permissions: Admins have Skill authoring access by default. To give a Contributor the same access, open Settings → Team, select the teammate, and enable Skill Authoring. This grants permission to create and update organization Skills without making the teammate an admin.

​
Load Flowglad MCP Skills into your agent

The Flowglad MCP server advertises io.modelcontextprotocol/skills during MCP initialization. This tells compatible clients that the server implements the agent Skill methods skills/list and skills/get. Each returned Skill includes skill:// resource URIs and content digests; clients can retrieve the files with resources/read. Connecting the server does not guarantee that a client will call skills/list or install the returned packages. Client support varies. Add a startup instruction to ChatGPT / Codex or Claude when you want the agent to refresh the Flowglad registry at the beginning of each session. MCP naming: The names are intentionally different. skills.list with a dot is a Flowglad tool for your organization’s versioned Skill records. skills/list with a slash is an MCP protocol extension method for discovering agent-facing Skill packages.

​
Add the startup instruction

Create the destination directory once before starting a new session, then add the matching instruction block below.

​
ChatGPT / Codex

Create the ChatGPT / Codex destination directory with mkdir -p ~/.agents/skills. Put the following block in ~/.codex/AGENTS.md to apply it to every project, or in a repository’s AGENTS.md to share it with that repository’s team.
AGENTS.md

​
Claude

Create the Claude destination directory with mkdir -p ~/.claude/skills. Put the following block in ~/.claude/CLAUDE.md to apply it to every project, or in a repository’s CLAUDE.md to share it with that repository’s team.
CLAUDE.md
You can ask ChatGPT / Codex or Claude to add the appropriate block to its own user-level instruction file. Review the proposed edit before accepting it because that file affects future sessions. The instruction is a best-effort agent bootstrap; if your environment must refresh Skills deterministically, implement the same synchronization in a trusted SessionStart hook or your MCP client integration.

​
Where Skills are used

Skills have two current consumers:
  • Compatible external agents discover and synchronize enabled packages through Flowglad MCP, then match them automatically or invoke them by name.
  • Automations pin an exact published Skill revision as an OWS skill-revision code resource. The Automation selects a package path and entrypoint; agent steps receive only the Skills authorized by the sealed run capability set.
The first path supplies reusable instructions to an external agent. The second supplies immutable code and supporting files to a repeatable Automation run. Only Skills available to the organization and Automation’s Space can be selected.

​
Creating and maintaining Skills

Open the organization menu → Skills → New Skill. In the Skill tab, enter Name, Slug, optional Description, and the reusable instructions in Markdown. Use the Files tab for any supporting package files, then select Create Skill. The Slug uses lowercase letters, numbers, and hyphens, begins with a letter or number, and is what users type after /. Creating or editing a Skill produces a draft. On the Skill detail page, edits autosave; select Publish Draft to verify and publish the draft. If verification fails, use the displayed diagnostics to correct the draft and publish again. Use History to inspect revisions, restore a superseded revision, or discard a mutable draft when those actions are available. Disable Skill removes it from external-agent discovery and Automation resolution without deleting revision history; Enable Skill makes it available again.

​
Create or edit a Skill with MCP

Use a connected external MCP client for package-first authoring. Call skills.get before editing an existing Skill. Use skills.create for a new package or skills.create_candidate for a revision based on the current active revision. Small packages can be supplied inline. For larger packages, call skills.create_package_upload_session, upload the requested objects, and finish with skills.finalize_package_upload_session. If verification was deferred, call skills.publish_revision. Poll skills.get_lifecycle_request for the accepted lifecycle request and its verification or promotion result. Publishing a Skill does not create, revise, or activate an Automation; use the Automation MCP lifecycle explicitly when one is needed.

​
Start from a Skill template

The Skills page includes Skill Templates. Search or filter the catalog, preview a template, and select Copy Skill. The copy becomes an organization Skill that can be edited and published like any other Skill.

​
Skills and Automations

A Skill defines how to perform reusable work. It does not decide when that work runs. Use an Automation when work needs to run manually, on a schedule, or in response to an event. An Automation can pin one or more Skill revisions so the same implementation is reused across runs.