Capacitor (working name)

Capacitor is the standalone system we are building to capture AI coding-agent sessions from the fleet, retain the raw evidence, enrich it into a usable corpus, and render that corpus in a console we own end to end. The inherited client is source-available; the server and web console are not inherited products.

Start here

PROMPT.md directs readers to start with Fleet. Read these in this order before changing a schema, a route, or a user interface:

1. reference/FLEET.md — the overriding product constraints: networked fleet capture, headless credentials, a machine dimension, and offline durability. It wins when it conflicts with another reference.
2. reference/SURFACE.md and the reference/ui-assets/ capture — the observed console, its session data, its navigation, and its visual system. The screenshots, tokens, CSS, icon map, and URL-addressable tabs are the acceptance reference for the web surface; they are not decorative inspiration.
3. reference/WAVES.md — its evidence method and gates are binding discipline; its proposed sequence is deliberately a hypothesis.
4. PROMPT.md — the outcome, acceptance criteria, and required feature-cut decision point.
5. reference/CROSS-REPO-SESSIONS.md and reference/REPEATING-CALLBACK-AUDIT.md — measured corrections that prevent false repository attribution and duplicate lifecycle facts.
6. reference/EVAL-DIRECTION.md — evaluation execution is core, while the inherited taxonomy is not.

The first integrated product contract is docs/SESSION-VERTICAL-SLICE.md. It starts at the captured Sessions surface and works backward through the required read model, normalization, and retained source evidence. It exists so that an API, schema, or parser cannot be treated as progress unless the real data it produces is rendered truthfully in the console.

docs/INDEX.md maps the complete documentation corpus without moving, renaming, or reducing it to summaries.

What is authoritative

The documents serve different jobs. Treat them accordingly:

Material	Role
LICENSE.md, NOTICE.md	legal and provenance boundary
PROMPT.md, then reference/FLEET.md	goal, acceptance criteria, and overriding fleet constraints
reference/SURFACE.md plus reference/ui-assets/	observed product, client contract, and visual/read-model authority
measured reference/ findings	factual corrections to any older design
reference/EVAL-DIRECTION.md	retained evaluation capability boundary
reference/WAVES.md	method and gates; not a prescribed implementation order
docs/schema/	derived design material to reconcile against the sources above before implementation
inherited docs/, kcap/, and npm/ material	implementation history and client-specific constraints, not a standalone product decision

A recovery branch, an unmerged pull request, a local stub, or a test database is evidence of work in progress. None defines the target product or proves that the standalone system exists.

Deployment boundary

No branch, pull request, fixture, or test database establishes a deployed Capacitor API or web console. Before making a present-tense deployment claim, take a fresh read-only Blood Arrow inventory and verify a real import plus browser render against the target. The selected PostgreSQL target for that remote acceptance is the isolated recovery/test database capacitor_test, not a production deployment.

The inherited client remains important: discovery, spooling, hooks, watchers, daemon behavior, and the wire payloads are source constraints to preserve while the standalone half is built. Vendor-hosted URLs, telemetry, tenant provisioning, update checks, feedback, marks, and typefaces are inherited references to replace or cut under the brief and licensing boundary.