[chrislacy]

User impact: Brave staff can ask their existing Agent to investigate, develop or verify Brave Android and receive scoped findings, captures and reproducible context without running a separate preparation framework.

The problem

BART focused on preparing automated QA cases through a CLI, frozen packages and a separate Agent runner. That added preparation overhead while the useful source-and-app verification workflow remained unfinished. It also limited the product to QA even though the same tools can support engineers developing and inspecting Android changes locally.

Reproduce

1. Start an Agent session in a Brave Core checkout with BART's skills made discoverable and a compatible Android target available.
2. Ask /brave-core-bart QA verify https://github.com/brave/brave-core/pull/40563. Save a local HTML report, captures and text-only context ZIP. Do not publish findings.

Observed before this change: BART offered separate case/run preparation skills requiring structured packages and CLI validation; the main development and verification entry point was not available.

Expected: the current Agent selects the needed source, build and device guidance, exercises the requested behavior and reports observations with visible evidence and clear coverage limits.

Initial local trials produced HTML reports and captures for Wallet system-theme restoration and Appearance settings on phone/tablet configurations. The later Appearance run explicitly reported a tablet-only missing icon/alignment difference; it also disclosed incomplete landscape interaction and transition evidence. These observations demonstrate useful initial coverage, not a complete regression pass or proof of causation.

The fix

This is a skills-first rewrite: one entry point for Brave Android work, one reporting skill and one optional reproduction skill, all followed by the Agent already in the session. Shared guidance supports autonomous execution and developer collaboration, current-Nightly QA, local changed-build checks, phone/tablet and orientation coverage, and foreground-preserving system-theme actions. Reports lead with a familiar QA summary and include privacy-safe evidence and a text-only context attachment when required. Optional local TOML configuration separates the working checkout and devices from QA/PR posting destinations.

The old preparation CLI, Agent launcher, schemas, package and package-dependent CI are removed outright. Useful source, build-identity and evidence knowledge is retained in maintained guidance. Local configuration and task artifacts remain ignored; development previews are explicitly distinguished from actual draft PR delivery.

Test plan

• git diff --check origin/main...HEAD - passed.
• Local document-link/anchor, skill-header, TOML and symlink checks - passed during implementation.
• Native local/personal skill discovery checks - found all three runtime skills, including from Core.
• Initial local Wallet and Appearance QA trials - reports and captures reviewed; observed results and coverage gaps retained privately.
• All 22 branch commits verify as signed; signing rewrite preserved the final tree and patches.
• Dedicated end-to-end feature-development and unattended local pre-submit demonstration - not yet performed.
• External report publication, actual draft PR delivery and independent live reuse of exported reproduction instructions - not yet demonstrated.
• CI passes cleanly - no replacement workflow is introduced; the deleted workflow depended on the removed runtime.