- Before changing userscript behavior or browser automation, read
docs/userscript-development.md. Use Ego's authenticated browser and ScriptCat as the default execution path; do not ask the user to repeatedly paste code into their userscript manager. - Inspect
node tools/scriptcat.mjs statusbefore starting a bridge. Reuse the running service when connected; do not start competing servers or install a duplicate copy of the same script. - Iterate through local edit → ScriptCat sync → installed-source SHA-256 check → real document reload → browser interaction → result verification. WebSocket delivery is not an installation acknowledgement. Sync explicitly per change; the bridge does not watch files.
- Add or extend repeatable browser regression coverage for changed features. Test the real userscript-manager sandbox and GM APIs rather than injecting source into the page as a substitute for installation.
- Check target identity and complete results, not just button visibility. For export features, validate actual downloaded files and remove temporary data. Test cross-origin frames and SPA navigation when relevant to the change.
- Use only the authorized Ego task space. Read Ego's installed browser skill before operating it, follow its ownership rules, and hand back necessary result pages when done. Never hard-code a user's space ID or profile.
- If the browser, authentication, or extension is unavailable, report the exact blocker and which checks remain unrun. Do not claim browser verification from syntax checks or a successful sync alone.
npm ci: install the bridge dependency at the repository root.npm run test:ego -- <space-id>: bridge, in-place update, source hash, real GM APIs, cross-site iframe, and modal fixture regression.npm run test:teams:entries -- <space-id>: Teams recap list, meeting selection, detail toolbar, and actual TXT/JSON downloads (Chinese Teams UI).npm run test:teams -- <space-id>: existing Teams recap transcript iframe.npm run test:atlassian -- <space-id>: real Jira title/link clipboard and navigation, linked Confluence page when available, and synthetic SPA/iframe regressions. Requires macOS; restores native clipboard formats in memory.npm --prefix tests cithennpm run test:smoke: existing Manebi smoke tests. These do not replace browser tests for other scripts.npm run test:pikpak -- <space-id>: native PikPak context menu and selection toolbar against a temporary loopback mock RPC, plus public-data pagination, nested paths, snapshot isolation, retry/cancel, and adapter failure checks. It restores RPC settings, never starts real downloads, and uninstalls its fixture.- Keep runnable bridge utilities in
tools/, browser tests and fixtures intests/browser/, and detailed instructions indocs/. Root README should link to the manual rather than duplicate it.
- Never commit credentials, cookies, control tokens, internal tenant URLs, meeting titles/text, employee identifiers, or browser dumps. Choose test meetings dynamically, without company-specific title fragments.
- Keep generated fixture scripts in ignored
tests/browser/.generated/. Validate real exports only in private temporary directories and clean them up even on test failure. Test output should contain status/counts, not content. - Before committing, inspect the exact staged files and diff for confidential identifiers and secret patterns; keep generated files and local service configuration out of the commit. Preserve unrelated user changes.
- Update relevant documentation and version/changelog for user-visible changes. Commit or push only when requested; report tests performed and remaining gaps.