[Robdel12]

Why

Comments and review decisions should disclose agent assistance while keeping the logged-in user as the author. Screenshot feedback also needs a precise destination: an agent looking at an image should be able to comment on a rectangle in that image’s pixels.

Approach

Detect documented Codex, Claude Code, Pi, Gemini CLI and OpenCode markers at the shared HTTP write boundary. Only static agent details are sent; marker values, session IDs and prompts stay local. Explicit --via-agent or --no-agent-attribution wins over VIZZLY_VIA_AGENT, then automatic detection. Conflicting markers stop the write with a clear error.

comment accepts --screenshot, --region x,y,width,height and --reply-to. Raw API comments and decisions use the same disclosure path. Review commands now use the shared credential selector and granted-scope checks, so explicit tokens take precedence and the selected credential is shown consistently. Help and bundled review guidance explain the new behavior.

Evidence

The full CLI suite passed 1,892 tests with six existing skips. Real subprocess/HTTP tests cover every supported harness, overrides, opt-out, conflicts, region validation, token precedence and read-only scopes. The built CLI also completed comment and decision workflows against a real authenticated server and database. Architecture checks, public type checks, lint and distributable builds pass.

Release after the server supports optional via metadata and screenshot comment regions.

[vizzly-testing]

Vizzly - Visual Test Results

CLI Reporter - Waiting for build

No builds received yet for this pull request.

CLI TUI - 3 changes need review

Result	Count	Review
Meaningful diffs	3	Required - inspect changed screenshots

Meaningful diffs needing review (3)

vizzly-help · 1202×1430 · 3.0% diff

vizzly-tdd-help · 1202×578 · 0.2% diff

vizzly-invalid-command · 1202×578 · 0.2% diff

Review changes

---

_{rd/agent-review-attribution · 192d8f9a}