prol-start-here

What it is for

Use it to start any Prolaborate task in your own signed-in session. You end with the identity the assistant is acting as confirmed, and your request routed to the right Prolaborate skill.

prol-start-here is the entry point for any Prolaborate task driven through the web interface. It does two things: it runs a short, read-only preflight so the AI knows whose session it is acting in, and it routes the request to the right Prolaborate skill. It also carries the working rules that every other Prolaborate skill inherits.

Capabilities it uses

When to use it, and when not

Use it first in any session that involves Prolaborate, before a more specific Prolaborate skill.

How it works

1. Check the sign-in. AI The AI opens the Prolaborate host and reads the page. If it lands on the login screen it stops and asks you to sign in. It never enters credentials for you. You How you sign in depends on the browser the AI drives:
   • The Claude Desktop browser pane starts with no sign-ins. You sign in to Prolaborate inside the pane.
   • Claude in Chrome uses your own Chrome, so an existing Prolaborate sign-in is already there. If you are signed out, you sign in in that tab.
   • VS Code's integrated browser. A page the assistant opens runs in an isolated session with no sign-in. You open Prolaborate in the integrated browser, sign in, and choose Share with Agent; the AI then works in that shared page.
2. Check for a live session token. (AI Power Tools, through browser control) A fixed script finds the session token in the page's own storage and narrows immediately to the access token. The scripts run inside the signed-in Prolaborate page, so the browser control has to be able to run a script in the page and return the result. A missing token means the application has not loaded in this page, so the AI opens the application root first. A stale token is normal and recoverable.
3. Read the identity. (AI Power Tools, AI interprets) The AI decodes the token's claims and reports who it is acting as. The claims that matter are the name (who the action is attributed to), the role (an administrator role means Super Admin), whether the account is read-only, and the groups (which drive Access Permissions). A Super Admin session is not a test of what a normal user sees, and it also means no authorization check stands between the AI and a mistake, so the skill asks for more care, not less.
4. Check the version. AI Power Tools The version appears in the page footer and on the health-check page. Each Prolaborate skill declares the version it was verified against (5.6.1.40). If the running version is newer, the AI prefers reading structured data over assuming screen layouts, and says which parts are unverified.
5. Check what the identity can see. AI Power Tools One read lists the repositories the identity can see. The AI confirms the intended repository is present and active. An inactive repository produces errors across Prolaborate that read like permission faults.
6. Route. AI The AI picks the skill from the table above, or declines a request the table does not route to a skill.
7. Apply the working rules. AI See below.

All five preflight checks are read-only. They run in order and stop at the first failure.

Working rules

These hold across every Prolaborate skill.

When something fails

The skill works through five causes in order before concluding a capability is missing:

1. The token aged out. Navigate to renew it, re-read the token, retry once.
2. Authenticated but not permitted. Check the role and your Access Permissions. This is a permissions answer, not a bug.
3. A refusal that says the user identity is empty: the AI holds a service token and not your own. It re-reads the token and does not substitute a registered application.
4. The repository is inactive.
5. The screen moved: compare the running version with the declared one and check the health-check page.

If none explain it, the AI states what it observed and stops. It does not try other ways to make the same change, because partial writes are visible to other people immediately.

What you get

• A short report of who the AI is acting as, the Prolaborate version, and the repositories visible.
• A routing decision, or a plain statement that the request is declined.
• Working rules applied to everything that follows in the session.

Tools it relies on

• Browser control (a built-in browser or Chrome), to open Prolaborate pages and run the preflight in the page
• The session token held in the page's own storage, read in the page and never copied out
• Prolaborate's repository list and its health-check page
• The shared Prolaborate session reference: Shared references

No server operation is called.

Sample prompts

1. Runs the preflight and reports identity, version and repositories.

2. Routes to prol-impact-analysis after the preflight.

3. Runs the preflight, then hands to prol-dashboards.

4. Declines the approval and does not act on it.

5. Declines the administration change and stops.

6. Reads the identity from the preflight and states it before any deletion, because a Super Admin session has no authorization check between the AI and a mistake. Deleting a dashboard is a consequential action, so the expected behavior is to list what exists and confirm which ones before deleting.

The application, repository and group names are illustrations. Replace them with your own.

Recommended models

Ratings are guidance based on the kind of reasoning the skill needs, not measured benchmarks.

Checking accuracy and completeness

1. Identity stated correctly. Compare the name and role the AI reports with the account shown in Prolaborate's user menu. Red flag: the AI reports an identity without decoding the token, or does not mention that an administrator role means Super Admin.
2. Token handling. Review the transcript for the token value, any part of it, or any read of the refresh token. None may appear. Red flag: a token in a message, a file or an address.
3. Repository list complete. Compare the repositories the AI lists with those visible in Prolaborate for the same user. Red flag: an inactive repository treated as usable, or a repository you can see that the AI omits.
4. Version reported. Compare with the footer. If the version is newer than 5.6.1.40, the AI must say the screen flows are unverified for it.
5. Routing. For each prompt above the AI picks the table's skill. Red flag: it acts in reviews or administration, or tries to create a review comment or a user "just to help".
6. Stop on login. Sign out in another tab and re-run. The AI must stop and ask you to sign in, with no credentials entered and no retries.

Limitations

• Verified against Prolaborate 5.6.1.40 on a Sparx-hosted tenant as a Super Admin.
• The access token expires after 240 seconds and the application renews it only on navigation. Long sequences re-read the token before each burst of calls.
• The server can end the session (concurrent-session blocking). The AI stops and does not retry.
• The skill depends on Prolaborate's current page layout. A newer Prolaborate can move screens.

See also Limitations and known issues.