Skip to content

B
bit-pm
切换分支/标签
  1. 01 5月, 2026 6 次提交
    • Vittorio's avatar
      支持连接器 · 427a845b
      由 Vittorio 提交于
      427a845b
    • Vittorio's avatar
      修改产品名称 · 99d0221f
      由 Vittorio 提交于
      99d0221f
    • Vittorio's avatar
      更换logo · d74517b0
      由 Vittorio 提交于
      d74517b0
    • Vittorio's avatar
      去掉广告，修改一些界面展示为中文 · d2955458
      由 Vittorio 提交于
      d2955458
    • keppo-bot[bot]'s avatar
      test: deflake setup and context E2E flows (#3319) · 30300641
      由 keppo-bot[bot] 提交于 
      ## Summary
- Deflake setup-flow E2E by clearing the fake OPENAI_API_KEY when a test
explicitly wants the setup screen.
- Update custom provider key setup to wait for the saved masked key UI
instead of the raw secret text.
- Wait for Smart Context settings persistence and give cloud sandbox
undo enough time to finish snapshot reconciliation.

## Root cause
Run: https://github.com/dyad-sh/dyad/actions/runs/25189808932

The red setup-flow failure was caused by two test assumptions drifting
from app behavior. The setup-screen fixture could inherit OPENAI_API_KEY
from an earlier test in the same worker, causing the setup banner to
disappear. The provider helper also waited for the full raw test API
key, while the app now saves and renders the key masked as test...2345.

Two retry-only flakes were separate E2E timing contracts: the Smart
Context test sent the next dump prompt before the off setting was
persisted, and cloud sandbox undo could still be syncing/restarting when
the digest poll hit the LONG timeout.

## Why this fix is correct
- Tests that opt into showSetupScreen now get an environment without the
fake OpenAI key, matching the setup-screen contract.
- The provider helper waits for the persisted saved-key state the UI
actually exposes, without asserting raw secret text.
- The Smart Context test waits on the settings file before sending the
prompt that depends on it.
- Cloud sandbox undo already uses EXTRA_LONG for preview startup;
applying the same budget to undo reconciliation matches the slower cloud
path seen in CI.

## Test plan
- npm run fmt && npm run lint:fix && npm run ts
- npm test
- PYTHON=/usr/bin/python3 npm run build
- PLAYWRIGHT_HTML_OPEN=never npm run e2e -- e2e-tests/setup_flow.spec.ts
- PLAYWRIGHT_HTML_OPEN=never npm run e2e --
e2e-tests/context_manage.spec.ts e2e-tests/cloud_sandbox.spec.ts

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------
Co-authored-by: 's avatarWill Chen <7344640+wwwillchen@users.noreply.github.com>
      30300641
    • keppo-bot[bot]'s avatar
      Deflake CI E2E failures (#3311) · 971aca34
      由 keppo-bot[bot] 提交于 
      ## Summary
- Deflake CI E2E failures from
https://github.com/dyad-sh/dyad/actions/runs/25126503338.
- Await streamed token-usage persistence before token-count invalidation
so the context-limit banner sees the real max token usage.
- Harden E2E helpers around GitHub sync completion, already-applied
proposals, queued Lexical submissions, visual-save button re-renders,
and Windows line endings.
- Update the component-tagger upgrade snapshot from 0.8.0 to 0.9.0.

## Root Cause
The failing run had several independent races and one stale snapshot:
- The context-limit banner could refetch token counts before the
assistant message's max token usage had been saved, leaving
actualMaxTokens null after a high-token stream.
- GitHub assertions snapshotted the connected-repo panel while the sync
button was still in its transient Syncing state.
- Some E2E helpers used one-shot Lexical fills/clicks where Playwright
can act during React or DOM replacement windows.
- The code-editor assertion compared raw file contents, which fails on
Windows CRLF output.
- The select-component upgrade snapshot still expected the older
component tagger version.

## Why This Fix Is Correct
- Awaiting the existing max-token update makes the later chat/token
invalidations observe the persisted usage instead of racing it.
- GitHub sync waits now use the stable connected-repo panel state:
Syncing hidden, Sync to GitHub enabled, and the panel-local success text
visible.
- Queued-message and visual-save actions now retry the exact UI contract
the tests require: the queued row appears or the save button is visible,
enabled, and clickable.
- The editor test normalizes only line endings, preserving the
file-content assertion while making it platform-independent.

## Test Plan
- npm run fmt && npm run lint:fix && npm run ts
- PYTHON=/usr/bin/python3 npm run build
- PLAYWRIGHT_HTML_OPEN=never npm run e2e --
e2e-tests/select_component.spec.ts --update-snapshots
- PLAYWRIGHT_HTML_OPEN=never npm run e2e --
e2e-tests/context_limit_banner.spec.ts e2e-tests/debugging_logs.spec.ts
e2e-tests/edit_code.spec.ts e2e-tests/github.spec.ts
e2e-tests/queued_message.spec.ts e2e-tests/visual_editing.spec.ts
- First run: 21 passed, 6 GitHub helper strict-mode failures due
duplicate success toast text.
- PLAYWRIGHT_HTML_OPEN=never npm run e2e -- e2e-tests/github.spec.ts
- npm test

🤖 Generated with [Claude Code](https://claude.com/claude-code)

<!-- devin-review-badge-begin -->

---

<a href="https://app.devin.ai/review/dyad-sh/dyad/pull/3311"
target="_blank">
  <picture>
<source media="(prefers-color-scheme: dark)"
srcset="https://static.devin.ai/assets/gh-open-in-devin-review-dark.svg?v=1">
<img
src="https://static.devin.ai/assets/gh-open-in-devin-review-light.svg?v=1"
alt="Open in Devin Review">
  </picture>
</a>
<!-- devin-review-badge-end -->

---------
Co-authored-by: 's avatarWill Chen <7344640+wwwillchen@users.noreply.github.com>
      971aca34
  3. 30 4月, 2026 2 次提交
  5. 29 4月, 2026 9 次提交
  7. 28 4月, 2026 5 次提交
  9. 27 4月, 2026 2 次提交
  11. 24 4月, 2026 1 次提交
  13. 23 4月, 2026 12 次提交
  15. 22 4月, 2026 2 次提交
  17. 21 4月, 2026 1 次提交
    • keppo-bot[bot]'s avatar
      Add deflake-e2e-from-run skill (#3247) · 1ab2dfc1
      由 keppo-bot[bot] 提交于 
      ## Summary

Adds a new skill, `dyad:deflake-e2e-from-run`, for root-causing E2E
failures from a specific CI run by reading the Playwright HTML report
artifacts directly — traces, screenshots, error-context, and app
stdout/stderr — instead of rebuilding and rerunning locally.

This complements the existing `deflake-e2e` and
`deflake-e2e-recent-commits` skills, which drive *discovery* by
repeat-running tests. This new skill is for when you already have a
failing run and want to root-cause fast.

## What the skill covers

- Downloading the merged `html-report` artifact with `gh run download -R
dyad-sh/dyad -n html-report`.
- `jq` queries against `results.json` for `unexpected` vs `flaky` test
buckets.
- Matching trace `.zip` hashes (paths in `results.json` are CI-side;
files are local in `/tmp/pw-report/data/`).
- Reading `test.trace` as JSONL to extract the step timeline.
- Correlating with app IPC logs that show up as `stderr`/`stdout` trace
events (gold for race-condition root-causing — e.g. `(proposal_handlers)
› IPC: get-proposal returned: …` at failure time).
- A short playbook of common failure shapes (disabled-button-after-fill,
navigation races, cross-test state) and their usual fixes.

Derived from the workflow used to root-cause #3246.

## Test plan

- [x] Skill file follows the same frontmatter format as existing skills
in `.claude/skills/`.
- [x] Verified via the use case in #3246 (full investigation done
end-to-end using exactly these steps).

🤖 Generated with [Claude Code](https://claude.com/claude-code)
<!-- devin-review-badge-begin -->

---

<a href="https://app.devin.ai/review/dyad-sh/dyad/pull/3247"
target="_blank">
  <picture>
<source media="(prefers-color-scheme: dark)"
srcset="https://static.devin.ai/assets/gh-open-in-devin-review-dark.svg?v=1">
<img
src="https://static.devin.ai/assets/gh-open-in-devin-review-light.svg?v=1"
alt="Open in Devin Review">
  </picture>
</a>
<!-- devin-review-badge-end -->
Co-authored-by: 's avatarWill Chen <7344640+wwwillchen@users.noreply.github.com>
Co-authored-by: 's avatarClaude Opus 4.7 (1M context) <noreply@anthropic.com>
      1ab2dfc1