Add verbose CLI progress events

[quanru]

Summary

• Add a global --verbose CLI flag that prints readable progress while Midscene commands are running.
• Make text-mode aiAct verbose output a clean human-readable timeline: start prompt, each plan's latest screenshot, planned reasoning, action details, action running/done status, timing, and final completion text.
• Surface aiAct screenshot paths next to the report under midscene_run/report/screenshots/, so agents and humans can open the exact screenshot referenced by each plan step.
• Preserve the generic verbose output for non-act commands and keep --verbose=jsonl as the structured machine-readable event stream.
• Reuse one CLI screenshot file writer for final tool-result screenshots and verbose dump screenshots; file extensions follow the image mime/format instead of forcing .png.
• Keep verbose payloads bounded: screenshot file paths are surfaced without base64, command args keep their nested shape, and dump updates use explicit minimal type guards.

Example

Real stdout from ./packages/web-integration/bin/midscene-web --verbose act --prompt "click the blue Search button" against packages/web-integration/tests/ai/fixtures/input-test.html:

[Midscene][aiAct] Start: click the blue Search button
[Midscene][aiAct][Plan 1/20] Thinking with the latest screenshot: midscene_run/report/screenshots/6b5f797f-d7fd-45b3-a5fa-d2f555ca3d5f.jpeg
[Midscene][aiAct][Plan 1/20] Planned: Click the blue Search button
[Midscene][aiAct][Plan 1/20] Action: Tap "the blue Search button" at (893, 364), bbox=(832,340,954,389)
[Midscene][aiAct][Action] Running: Tap at (893, 364)
[Midscene][aiAct][Action] Done: Tap cost=891ms
[Midscene][aiAct][Plan 2/20] Thinking with the latest screenshot: midscene_run/report/screenshots/068dd6e3-8296-41ce-9494-95d74e6f3747.jpeg
[Midscene][aiAct][Plan 2/20] Planned: The user's instruction was to "click the blue Search button", and the previous action has already executed that click. The current screenshot shows the same interface with the S...
[Midscene][aiAct] Complete: The blue Search button has been clicked.
Action "act" completed.
Result: The blue Search button has been clicked.

Validation

• pnpm exec biome check packages/shared/src/cli/verbose.ts packages/shared/src/cli/screenshot-file.ts packages/shared/src/mcp/tool-generator.ts packages/core/src/types.ts packages/core/src/agent/tasks.ts packages/shared/tests/unit-test/tool-generator.test.ts
• pnpm --dir packages/shared exec vitest run tests/unit-test/tool-generator.test.ts -t "verbose"
• pnpm exec nx test @midscene/shared
• pnpm exec nx build @midscene/web
• Real midscene-web run against local fixture: ./packages/web-integration/bin/midscene-web --verbose act --prompt "click the blue Search button"; verified stdout includes Start, per-plan screenshot paths, Planned, Action, Running, Done, and Complete lines.
• pnpm run lint was run, but it fails locally on unrelated untracked files packages/web-integration/swipe-verify.cjs and packages/web-integration/swipe-verify2.cjs with lint/style/useSingleVarDeclarator; those files are not part of this PR.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 66136ee73c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Redact sensitive CLI args from verbose events

When --verbose is used, the command_start event serializes all handlerArgs directly; for the computer RDP tools the accepted init args include password (packages/computer/src/mcp-tools.ts), so a command such as midscene-computer --verbose act --host ... --username ... --password ... prints the password in the structured JSON progress stream. Because these verbose logs are intended for progress collection and may be persisted by CI or wrappers, sensitive keys such as password/token/apiKey should be redacted before emitting the args payload.

Useful? React with 👍 / 👎.

[cloudflare-workers-and-pages]

Deploying midscene with    Cloudflare Pages

Latest commit:	7ece567
Status:	✅  Deploy successful!
Preview URL:	https://06d3ebb4.midscene.pages.dev
Branch Preview URL:	https://feat-shared-cli-verbose-prog.midscene.pages.dev

View logs