Jobs and keys

Security scans need no key. Drafting skills from sessions uses the workspace's model provider on the workspace's own key, and the provider bills the calls.

Two jobs

Decision model training is a separate, gated pilot and is not described by this page. See Skills and review or Draft a skill from your sessions.

The model key

The hosted service uses the workspace's model provider and takes no other provider's key. An admin adds the key on the Keys page. Draft jobs use the models configured for the hosted release, using this key. This page does not establish which model or pricing tier a particular deployed job used. You pay the model provider. Slancha does not resell model access.

If the deployment names its job model, saving asks the model provider whether the key can read that model. A successful answer confirms only that read, not that drafts will be allowed to run. If the provider answers 401, 403 or 404, the key is not saved. If no model is named, the provider cannot be reached, or it answers with any other status, the key is saved unchecked and you are told so. The key is encrypted when you save it. It is never shown again: you can replace it or delete it, not read it. Only an admin can change it. If you delete it, draft jobs are refused until you add one again. A key for another provider, saved before the hosted service moved to its current model provider, is listed only so you can delete it. No job uses it.

Jobs

Jobs shows the latest 50 runs, newest first. Filter by skill. After queued and running, a job ends as:

• succeeded
• failed
• cancelled

Each job shows its skill, version, starter, timestamps, status and estimated cost. Running jobs show progress. Completed jobs do not change.

Scans show the verdict and findings. Drafts show session-label counts and link to any proposal. Full logs are not available in the console.

An admin starts drafts. An agent token from slancha auth --manage can start drafts and outcome jobs too, while an admin allows it in Workspace settings. A proposer or above can cancel a running job. A cancelled scan leaves the version unapproved until a scan finishes clean. Only one draft runs for a skill at a time, and a skill with a change waiting for review cannot start a draft.

Cost and the cap

A scan costs nothing. A draft costs whatever the model provider charges your key. The job page shows an estimate from list prices.

Each run has a server-configured cost cap that cannot currently be changed in the console. Runs stop after exceeding it. Prices, retries and long runs can cause overruns; the cap is not a guaranteed spending limit.

Review evidence

Hosted Slancha does not create eval sets, compare skills with a baseline, or rank versions. Slancha shows scan findings and usage counts; it does not grade skill quality. Publishing requires a clean scan and human approval.

What you see instead:

• the security scan, with each finding;
• the diff, and the sessions a draft was written from, with their labels;
• how often each version was used, and how often it succeeded or failed, as the plugin reports it (tool invocation success, not task quality);
• the workspace's ledger of every draft and how it ended.