Skills and automated work¶

Outcome¶

You will understand when Threadory runs AI work automatically, where its results appear, and how an administrator can start an eligible skill manually.

Before you begin¶

Every project member can inspect outcomes in Approvals, Activity, and Reports.
The host-wide Skills catalog and manual controls are available to project administrators and owners.
Automatic skills need the relevant connected source and fresh synced data.

Understand a skill's trigger¶

Open Skills.
If you are a member or viewer, follow the links to Approvals, Activity, and Reports. These are the day-to-day result surfaces.
If you are an administrator or owner, open a catalog card with Details.
Read Trigger and Creates approvals before relying on the skill.
Check Recent runs for the current project.

Skills are installed for the Threadory instance, not separately for each project. A catalog card therefore describes availability on the instance; it does not prove that this project has the connector or data needed for a useful run.

The catalog distinguishes:

Automatic — starts from a source event, connector sync, incoming message, or schedule. Run now is disabled in the catalog.
On demand — an administrator or owner can start it with Run now.
entity-specific skills — may run automatically or from the relevant entity context, even when a bare catalog run would not have enough input.

Start an on-demand skill¶

Open an On demand skill and read its description.
Confirm that the active project contains the required data.
Select Run now.
Confirm the dialog.
Select View run in Activity after Run started appears.
Follow the run until it reaches a terminal status, then inspect its output in Approvals, Reports, Knowledge, or the connected source.

Starting a run only confirms that it was queued. It does not mean the skill completed successfully.

Review automatic results¶

Open Activity after a connector sync or expected trigger.
Open the run and check its status and summary.
Resolve any proposed external write in Approvals.
Check extracted decisions, action items, digests, or topics in Reports and Knowledge.
Verify external comments, issues, or replies in their source system.

Review your open merge requests right after connecting¶

A new GitLab or GitHub connection does not make you wait for the next push. Right after the connector is installed, Connectors (and the onboarding screen you return to) offer the card Review your last open merge requests now.

Select Review now.
Threadory takes up to three of the most recently updated open merge or pull requests and starts a review for each. The card lists them with links to their runs in Activity.
Each review is posted on the merge request like any other review, usually within a few minutes.

Any project member or above can start it. It can be used once per connector every ten minutes; a second attempt within that window is refused with a “try again in a few minutes” message. A merge request whose current head was already reviewed is not reviewed twice. If nothing is open, the card says so and the next merge request you open is reviewed automatically. A Free project that has already used its monthly review allowance (see below) is refused with a message saying so, since nothing would be reviewed.

Review limit on the Free plan¶

The Free plan includes 100 posted reviews per project per calendar month. Each review that Threadory posts counts as one; the counter restarts on the first day of the month (UTC). When the limit is reached, Threadory does not review new heads and instead leaves one short comment on the merge request, once per head, saying that the monthly limit was reached and linking to the plans. A project that requires human approval of every comment (auto-post off) gets no such comment; its merge requests are simply not reviewed until the limit lifts. Trials, Team and Enterprise have no review count limit. Upgrading takes effect immediately for the next review. A head that was skipped because of the limit is picked up again by the periodic sweep once the allowance is available (a new month, or an upgrade), or sooner when it is pushed again.

Review comments that Threadory posts to a merge request or pull request, and the weekly digest sent by e-mail or Telegram, can end with a short signature that links to Threadory:

A review reads “Reviewed by Threadory with project context: …”. The counts come from the context Threadory linked to the change (at most 12 items are counted). It also states that the review is AI-generated and that a human still decides.
A digest reads “Powered by Threadory”. The copy stored in Reports is not signed, only what leaves the product.
An issue that Threadory creates in your tracker from an action item ends with “Created by Threadory” and a link (a plain address in trackers that do not render links).
The public page of a shared document (/d/<token>) ends with “Shared with Threadory · Specs with client comments and sign-off · Start free”. The link goes to the Threadory site and never contains the document link. Without the footer the page keeps only a plain “Shared with Threadory” credit.
A PDF of a document has “Exported from Threadory · threadory.ai” at the foot of every page.

On the Free plan the footer is always added. On a paid plan outputs are white-label by default. An owner or admin can turn the footer on in Settings → General → Threadory footer on shared outputs; the setting is ignored on Free.

The invitation e-mail Threadory sends to a new member always names who invited them and links to the Threadory site; it is not affected by the setting.

Verify the result¶

Activity contains a run for the expected skill and project.
The run reaches a completed status rather than remaining queued or running.
Any approval or external write matches the run summary and source context.

Common problems¶

Symptom	Likely cause	What to do
You see “How AI helps” instead of the catalog	Your role is not admin or owner	Use the linked outcome pages or ask an administrator to inspect the catalog
Run now is disabled	The skill is automatic or needs entity context	Read Trigger and create the expected source event
A run starts but produces no useful result	Required source data is missing or stale	Sync the relevant connector and inspect the skill description
No automatic run appears	The connector, event, or schedule has not fired	Check the connector and Activity, then reproduce the trigger safely
A run remains queued or fails	The workflow service or an upstream dependency is unavailable	Retry later and give the run link to your operator if the failure persists
Output is waiting in Approvals	The skill proposes an external write	Review its source context; do not bypass the approval merely to clear the queue

Next step¶

Use Approvals for proposed writes, or connect the project to an MCP client.