Skip to main content

​
Pilot Onboarding

This page is for pilot operators running Prism today, before Prism Cloud launches. It’s the short version — what you sign up for as a pilot, what’s awareness-only, and where to go for operational depth. Pilots run a Local-or-LAN install (see INSTALL). Once you’re up, Prism’s customer-care lane is wired in by default: questions go to Sage, incidents go to Sage → Clara, and there’s an opt-in per-incident channel that lets the Prism team see Prism-product defects so we can act on them.

​
Two ways the customer-care lane reaches you

Helper-mode questions — info-class intake. Ask Sage anything about Prism (CLI usage, config, SPECs, ADRs, method fragments, error messages, agent roster). She answers in-thread from Prism memory and verbs. Nothing leaves your install. See Customer Service Agent for the full walkthrough. Incident-mode workflow — incident / enhancement / recommendation intake. When something is broken or you have a defect to report, Sage routes the case to Clara, who triages, classifies severity / scope / kind, and prepares a redacted packet for operator review before send. If you approve, Clara’s ET helper transports the packet via gh issue create to the Prism team’s tracking repo. See Incident Management for the full flow, classification rubric, failure modes, and bridge-routing details.

​
Customer awareness — what to know before you opt in

Three things to be explicit about. None of them require action right now; they’re for awareness so you know what you signed up for.

​
1. The destination repo is private and invite-only

When phone-home fires, it creates a GitHub issue on the private repo FrankTewksbury/Prism via gh CLI. The Prism team has to invite your GitHub account as a collaborator on that repo (Triage or Write access — enough to create issues) before prism_phone_home works on your machine. Without the invite, the verb fails with phone_home_repo_access_denied and the incident stays local. Verifying you have access: 
gh auth status                                          # signed in as the invited account
gh repo view FrankTewksbury/Prism > /dev/null && echo OK # has access
If you intend to use the phone-home channel and don’t have the invite yet, ask the Prism team. If you don’t intend to use it, ignore this step — Sage’s info lane and local-only incident triage still work.

​
2. The repo may not stay private forever

Treat every incident packet you approve as if it could become public someday. Redact under that assumption. Clara prepares a packet with PII removed, customer content removed, secrets/tokens stripped, and hostnames / paths / tenant names reviewed — but the final review is yours. If the redacted packet still contains something you wouldn’t be comfortable seeing in a public issue, decline approval or ask for further redaction before approving.

​
3. There is an opt-out path

Three valid opt-outs, all of which keep local accounting intact (postmortem + journal + reporter signal) — opt-out means “don’t ship the packet out of the install,” not “don’t track the incident”:
  • Per-incident decline. Clara surfaces the rendered packet; the operator declines. Recorded locally as operator_rejected.
  • Scope-based. customer-internal scope never phones home (rejected at the schema layer); mixed scope phones home only the redacted Prism-product subset.
  • Operator never approves anything. prism_phone_home requires operator_review.approved == true — absence of approval is the channel-level opt-out. Sage and Clara keep classifying and recording; ET just never ships.
Dry-run path: invoke prism_phone_home(dry_run=true) to see the rendered title / body / labels before any GitHub call. Skips operator-review + redaction-report enforcement so it’s safe to invoke without ceremony — useful for getting a sense of what the channel would expose.

​
v0.1 limitations, in summary

All of these ship resolved in later versions; surfacing them so pilots know what’s on the roadmap:
  • Per-operator gh auth. v0.2 swaps gh CLI transport for a Prism-Bot GitHub App; no per-operator auth then.
  • Private invite-only destination. The invitation-driven onboarding flow (operator-invited new pilots get a pre-keyed installer) ships with the cloud + admin-console lane (next big Prism push).
  • One-way channel. Today phone-home is send-only — operators see issue comments by checking GitHub, not by Prism surfacing them back into Clara. A future gh_issue_sync scheduled job will close that loop.

Customer Service Agent

info-class help: ask Sage anything about Prism or your project, get answers in-thread from memory + verbs.

Incident Management

The full operational flow when something is broken: Sage classify → Clara triage → operator review → ET ship.

INSTALL — full reference

Every flag, preflight, and troubleshooting case for the installer. Pilots run the Local or LAN backend mode.

LAN install walkthrough

Connecting a pilot machine to a shared Prism backend on the local network.
Last modified on May 18, 2026