Hot Update
This page exists so presales, delivery, training, and customer discussions can all quote the same explanation.
ExecFabric already supports customers continuing to add or update script-based capabilities without redeploying the whole platform. But that path is still a controlled update, not an open self-serve release flow.
What It Means
What It Is Not
Hot Update Flow
Prepare the folder according to the agreed structure. At minimum it should include the entry script and README.md. The current default entry files for Python and Shell text scripts are main.py and main.sh.
The customer places the script into their own tenant directory. Different customers cannot mix directories, and public/ is maintained only by the platform.
The platform confirms which tenant the script belongs to, whether AI can call it, whether manual execution is allowed, and whether manual confirmation is required.
The runtime uses controlled relative paths rather than arbitrary absolute paths outside the managed script workspace.
The platform completes script registration, script-version registration, Skill association, Skill-version registration, and tenant authorization binding.
The platform verifies that the capability is visible and executable in the current tenant, invisible elsewhere, and not leaking across AI recommendation boundaries.
Hot Update Submit
| Submitted item | Required | Purpose | Notes |
|---|---|---|---|
| Script files | Required | The real execution content | At minimum include the agreed entry file. Today the default for Python and Shell is main.py or main.sh. |
README.md | Required | Used for platform review and later maintenance | It should explain purpose, input, output, prerequisites, and whether files are generated. |
| Owning tenant | Required | Determines the final authorization scope | Different customers cannot mix directories or authorization. |
| Business-scenario description | Required | Helps the platform judge naming and integration method | It should explain what problem the script solves and who will use it. |
| Whether AI can call it | Recommended | Determines whether the intelligent brain may recommend it | Not every script enters AI recommendation scope by default. |
| Whether manual execution is allowed | Recommended | Determines whether the page opens a manual entry | This can be controlled separately from AI invocation. |
| Whether manual confirmation is required | Recommended | Controls the confirmation step for higher-risk actions | This fits capabilities that rewrite data or trigger external actions. |
Why It Matters
Misunderstanding
FAQ
No. After the script update is complete, it is still registered and authorized only inside the tenant it belongs to. Other tenants do not automatically gain visibility or execution permission.
Not by default in the current 1.0 scope. The platform still keeps the review, registration, authorization, and permission-check flow and does not support direct formal release around the platform.
No. Recommendation scope is constrained jointly by tenant and permission. AI does not recommend another customer's script or capability across tenants.
Yes, but the usual path is not to let the browser touch a local path directly. It is to use execfabric-cli and the Local Agent direction as the bridge for local execution.
Next Read
During formal delivery evaluation, it is usually more useful to read customer flow, the file path, and the onboarding checklist together than to discuss hot update alone.