> ## Documentation Index
> Fetch the complete documentation index at: https://prism.ntecdev.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Customer Service Agent

> Ask Sage anything about Prism or your project — config questions, CLI/verb help, ADR lookups, error explanations, agent roster, project status. Sage answers directly from Prism memory and verbs, no incident triage required.

# Customer Service Agent

If you have a Prism question — how does this verb work, where's the config for X, what does this error mean, what's the spec/ADR that covers Y — send it to **Sage**. Sage is the Customer Care intake desk, but the intake is not just for incidents. The `info` classification path is the conversational help path: Sage answers from Prism memory directly, no Clara involvement, no operator review, no phone-home.

Use [Incident Management](/incident-management) when something is broken or you suspect a Prism-product defect. Use this page when you want an answer.

## How to ask

```bash theme={null}
prism_signal --to-identity Sage \
             --signal-type Question \
             --payload '{"summary": "How do I rotate the API key on a LAN install?"}'
```

Or, more commonly, from inside an agent session that's already wired into the Prism MCP:

> Hey Sage, what is the recall-class split in ADR #58?

The customer agent (Jade in this example) signals Sage. Sage classifies the request and, if the classification is `info`, replies in-thread.

## What kinds of questions Sage handles

Anything that resolves to a memory lookup, a verb explanation, a config pointer, or a doc cross-reference. The full SPEC-114 intake rubric is in the [Sage and Clara persona contracts](/specs/spec-114-v0-1-sage-and-clara-customer-care-incident-personas); the table below is the operational view from a customer agent's perspective.

| Category              | Example question                                  | What Sage does                                                                                 |
| --------------------- | ------------------------------------------------- | ---------------------------------------------------------------------------------------------- |
| **CLI / verb usage**  | "How do I run `prism_release_cut`?"               | Cites the verb signature + the release-runbook entry; quotes a usage example.                  |
| **Config lookup**     | "Where do I set `PRISM_GIT_VERSION`?"             | Cites `templates/CLAUDE.md` BIOS conventions + the build-arg chain (PR #341).                  |
| **Spec / ADR**        | "What's the recall-class split in ADR #58?"       | Looks up the ADR body, summarises the rule, links the canonical home.                          |
| **Error explanation** | "What does `phone_home_repo_access_denied` mean?" | Resolves the failure-mode code to its SPEC anchor + remediation.                               |
| **Roster / who**      | "Who handles install on Windows?"                 | `prism_whois`-style lookup; cites Cherry (Windows persona) or Lafonda (install lane).          |
| **Project status**    | "What phase is PID-PGR01 in?"                     | Reads `prism_status` / project metadata, cites phase + last wrap.                              |
| **Memory recall**     | "Have we seen this stack trace before?"           | `semantic_recall` over postmortems + journals; cites matching prior incidents.                 |
| **Method fragments**  | "What's the Wallbreak protocol?"                  | Cites `method.wall-break.persistence` fragment + the relevant agent-operating-doctrine pieces. |

If the question doesn't resolve cleanly to one of these — for example, the question is actually a defect report — Sage reclassifies and hands off to Clara per the [Incident Management](/incident-management) flow. The customer agent doesn't have to know in advance which path applies; that's what Sage's classification step is for.

## What Sage can and cannot do

Sage reads. Sage does not write to your customer project. Specifically:

* **Sage reads** Prism memory (`semantic_recall`, `prism_memory_cite`), SPEC and ADR bodies (`prism_spec`, `prism_governance_lookup`), method fragments, agent roster (`prism_whois`), project metadata (`prism_status`, `prism_context`), and the canonical docs surface.
* **Sage does not** edit your project files, run code in your repo, mutate your project's `.prism/` metadata, write durable artifacts into your project's PID, or modify your agent roster.
* Durable artifacts Sage produces — care intake records, classification journals — live in Sage's home project (`PID-PGR01`), not in your customer project. See the [Bridge Agents](/incident-management#bridge-agents) routing contract for the full artifact-ownership rule.

If you want a project-local artifact (a `prism_todo` in your project, a `prism_journal` entry under your PID), Sage will suggest the verb and you run it yourself. That's `info` mode in a nutshell: Sage informs; the customer agent acts.

## Cross-project reachability

Sage is a **bridge agent** (`persona_class='bridge'` per [SPEC-116](/specs/spec-116-v0-1-bridge-agents-cross-project-customer-care-routing)), which means she's reachable from every project in the tenant. You do not need to install Sage in your customer project. `prism_signal(..., to_identity='Sage')` resolves through the bridge routing path after same-project resolution fails. Same-project personas with the name "Sage" (unlikely but possible) take precedence over the bridge, so local-name shadowing is impossible.

The same is true for Clara, but Clara is the [Incident Management](/incident-management) path — don't address her directly for `info` questions; let Sage classify.

## What Sage doesn't replace

Sage is a help desk, not a documentation site. If you want to read the canonical reference for a verb, a SPEC, an ADR, or a method fragment, go to the page itself. Sage points; the docs answer in depth.

Sage also doesn't replace the lane owner. For substantive design questions, route through the lane owner directly:

* **Architecture / SPEC questions** — Texi.
* **Engineering / implementation** — Donna.
* **Governance / methodology / vocabulary** — Candi.
* **Install / scaffold** — Lafonda.
* **Release / cutover** — Samantha.
* **Docs / Mintlify surfaces** — Desiree.

Sage is happy to tell you which lane owner to ask. She just won't override their judgment with her own.

## Related

<CardGroup cols={2}>
  <Card title="Incident Management" icon="user-doctor" href="/incident-management">
    The other half of Sage's intake — what happens when the classification is `incident`, `enhancement`, or `recommendation` instead of `info`.
  </Card>

  <Card title="Pilot Onboarding" icon="rocket" href="/pilot-onboarding">
    Customer-awareness setup for pilots: invite-only repo, redaction posture, opt-out paths.
  </Card>

  <Card title="SPEC-114 — Sage and Clara" icon="user-pen" href="/specs/spec-114-v0-1-sage-and-clara-customer-care-incident-personas">
    The persona contracts and Sage's full intake-rubric semantics.
  </Card>

  <Card title="SPEC-116 — Bridge Agents" icon="bridge" href="/specs/spec-116-v0-1-bridge-agents-cross-project-customer-care-routing">
    Why Sage is reachable from every project without per-project install.
  </Card>
</CardGroup>
