Sync to the Claude Skills API
Send releases of a repository to your own Claude Skills API workspace with your own key. Set a sync target, start a sync, and read the five outcomes, including what the destination refuses.
If your assistants load skills from the Claude Skills API, SKaaS can keep that copy in step with a repository. It uploads a release with your own API key, so the skills land in your own workspace. SKaaS keeps the record of what it sent.
Edition and plan: every edition and every plan. You need your own key for the Claude Skills API.
How do I set it up?
Give the repository a sync target. You need the repos:admin scope. The key is stored sealed and is never shown again, only its last four characters.
curl -sS -X PUT "https://api.skaas.io/v1/repos/$REPO_ID/claude-sync/target" \
-H "authorization: Bearer $SKAAS_TOKEN" -H "content-type: application/json" \
-d '{ "api_key": "'"$CLAUDE_API_KEY"'", "workspace_id": "wrkspc_example" }'
workspace_id is for a key that can act on more than one workspace. Send "enabled": false to pause the target. DELETE on the same address removes the key. GET /v1/repos/$REPO_ID/claude-sync shows the target and every skill that was synced, with the remote skill, the remote version and the release hash that was sent.
Where do I do this in the dashboard?
Open the repository and choose Integrations. The section Sync to the Claude Skills API has a form that sets the key and the workspace, and it shows the key by its last four characters. Once a key is set, the section shows a form to start a sync, with the outcome of each skill, and a form that removes the key. Each outcome is one of the five statuses below. A deployment that has not set the sync up shows "Sync to the Claude Skills API is not set up on this deployment". The key is sealed with SKAAS_ENCRYPTION_KEYS, so the operator must have set it and run a server that has the sync service.
How do I start a sync?
A sync is started for one skill at a time, and it sends one release: the head, or the one that you name.
curl -sS -X POST "https://api.skaas.io/v1/repos/$REPO_ID/claude-sync/syncs" \
-H "authorization: Bearer $SKAAS_TOKEN" -H "content-type: application/json" \
-d '{ "skill_id": "'"$SKILL_ID"'" }'
The request succeeds whatever the outcome is. Read the status in the answer.
| Status | Meaning |
|---|---|
synced |
The release was sent. created means a new remote skill, and versioned means a new version of an existing one. |
unchanged |
The remote copy already holds this release. Nothing was sent. |
skipped |
There is no target, the target is paused, or the artifact is not a skill. |
rejected |
The destination would refuse the package. The answer lists each violation with its path. Nothing was sent. |
failed |
The destination did not accept it. The answer says whether a retry can help and when, and gives the request id. |
A transient failure is retried up to three times before it is reported. Two syncs of one skill never run at the same time.
What does the destination refuse?
SKaaS checks the package against the rules of the Claude Skills API before it sends anything, and it changes no file. The name has 1 to 64 characters of lowercase letters, digits and hyphens, and it holds neither claude nor anthropic. The description is not empty, has at most 1,024 characters and holds no XML tags. The package has one SKILL.md, at its root, so rename a nested one to references/<topic>.md. All files together are at most 30,000,000 bytes uncompressed.
Run skaas audit --target claude-skills-api on your machine to see the same problems before you publish. The audit and the export spell this destination differently. The audit calls it claude-skills-api and the export calls it claude-api. Both mean the same thing.
What does a sync not do?
It does not sync on publish by itself, it does not remove a remote skill when you archive one here, and it does not upload to the Claude apps. For the apps, export a .skill bundle with skaas export --format skill --target claude-apps and upload it by hand.