Wiki

In-Depth User Guide

Detailed guide to requests, agent sessions, reviews, approvals, promotions, and deeper product behavior.

Repository View Source

In-Depth User Guide

1. Working with Requests

Requests are the core governed object in RGP.

On a request detail page you can inspect:

  • overview and status
  • runs
  • checks
  • reviews
  • promotions
  • artifacts
  • agent sessions
  • timeline and event history

Request Statuses

Common statuses include:

  • Draft
  • Submitted
  • Validated
  • Planned
  • Queued
  • In Execution
  • Awaiting Input
  • Awaiting Review
  • Approved
  • Promotion Pending
  • Promoted
  • Completed
  • Failed

2. Agent Sessions

RGP supports assigning a request directly to an integrated agent.

How It Works

  1. open a request
  2. navigate to the agent surface
  3. choose an agent integration
  4. start a session
  5. interact across multiple turns
  6. accept the response when you want the request to continue

Important Behavior

  • the session persists transcript and state
  • agent output can stream live
  • the request can pause in awaiting_input
  • accepting the result resumes workflow

For governed external runtime sessions such as sbcl-agent, the session page also becomes a runtime-control surface.

Operators can now:

  • inspect governed runtime environment, thread, and turn references
  • see pending approval checkpoints exposed by the external runtime
  • approve or resume specific runtime work-items
  • import runtime-produced artifacts into RGP as first-class governed artifacts with lineage

That means an sbcl-agent session is not only a conversation transcript. It is a governed view over a durable external runtime.

This is intentionally different from a normal chatbot. The session is part of the governed work record.

3. Reviews and Approvals

Requests that require review move into the review queue.

Reviewers can:

  • approve
  • reject
  • request changes
  • use reassignment / override flows where allowed by governance

Promotion approval happens separately and remains explicit.

4. Promotions and Deployments

Promotion is the governed finalization step for work that must be formally advanced.

A promotion may include:

  • preflight checks
  • approver assignment
  • deployment execution
  • completion recording

Requests are not supposed to silently promote or complete without visible governance evidence.

5. Analytics

RGP includes several analytics families:

  • workflow analytics
  • agent analytics
  • delivery analytics
  • performance analytics
  • cost analytics
  • portfolio and organizational summaries

Most analytics pages support filtering by:

  • portfolio
  • team
  • user
  • time window

Several also support comparison views and time-series charts.

6. Admin Functions

Templates

Use Admin → Templates to manage template identities and versions.

The drill-down workbench supports:

  • draft creation
  • field design
  • conditional rules
  • routing rules
  • governance requirements
  • validation
  • preview
  • comparison
  • publish / deprecate

Organization

Use Admin → Organization to manage:

  • teams
  • memberships
  • users
  • portfolios

The org page is built around a hierarchical team-membership view and drill-down management pages.

Integrations

Use Admin → Integrations to manage:

  • runtime and external system bindings
  • direct-assignment agent integrations
  • provider settings such as model, base URL, workspace, and secret rotation

Secrets are managed through the integrations surfaces but are intentionally not readable back in cleartext.

7. Operational Queues

Operational users should rely on the queue pages for active work:

  • Review Queue
  • Blocked Requests
  • SLA Risk
  • Promotion Pending

These are table-driven and intended for sorting, filtering, and action.

8. Troubleshooting

Request Is Stuck

Check:

  • status
  • blocking checks
  • pending review items
  • promotion state
  • agent session state

Agent Session Is Not Progressing

Check:

  • the selected integration
  • session state
  • whether the latest turn completed
  • whether the request is waiting on human input
  • whether a governed runtime approval checkpoint is blocking execution
  • whether the runtime artifact or approval action you expect has already been reconciled into the session page

Analytics Look Incomplete

Check whether the relevant request/run/review/promotion activity actually occurred. Analytics are derived from governed work records and performance telemetry.

9. Mental Model for Users

The most important thing to understand is:

  • RGP is not mainly a chat app
  • RGP is not mainly a dashboard
  • RGP is a governed work system

If you think in terms of request → governed execution → review → promotion → analytics, the product will make sense quickly.