Inspect Kiln’s interface in a real browser, capture screenshots, and run or debug end-to-end tests that check complete user flows. A sandbox with sample project data provides populated screens for checking UI changes.
中文简介正在准备中。当前显示英语简介。
Look at Kiln's UI in a real browser, and run its end-to-end tests. Use when checking UI you are changing, taking a screenshot of the app, driving the app with playwright-cli, starting the dev sandbox with playwright_server.sh, running or debugging `npm run tests:e2e`, or working with the seeded fixture project that gives the app's screens data to render.
安装前可以查看交给 AI 代理的指令内容。
Two separate things share one browser install:
playwright-cli — a browser you drive from the shell, for looking at the UI
you are changing and taking screenshots of it. This is what you want when
building UI.npm run tests:e2e — the end-to-end suite, in app/web_ui/tests/e2e.If neither works, the install is probably missing: bash .config/utils/setup_env.sh --add-playwright adds the browser and playwright-cli (~800 MB together). See that
script's --help for what a cloud environment needs.
Start a server, open the app, and land on a page with data in it:
.agents/scripts/playwright_server.sh start # prints http://localhost:6544
start prints the next three commands with the seeded ids already filled in. Run
all three, in that order — the app redirects to a task picker without the
localStorage state, whatever URL you ask for:
playwright-cli open http://localhost:6544
playwright-cli localstorage-set ui_state \
'{"current_project_id":"<id>","current_task_id":"<id>","selected_model":null}'
playwright-cli goto http://localhost:6544
Then goto any deep link you like. Run .agents/scripts/playwright_server.sh status
if you are not sure the server is running.
When you are done, close the browser and stop the server:
playwright-cli close
.agents/scripts/playwright_server.sh stop
The sandbox keeps its data in app/web_ui/.agent_dev_home and is seeded from a
committed fixture project, so the screens have content instead of an onboarding
wizard. It never touches real Kiln projects.
If OPENROUTER_QA_KEY is set in the environment, seeding writes it into the
sandbox's settings and the app comes up with OpenRouter connected, so live model
calls work. start reports it when it is there — assume nothing is connected
until it says so.
When it is connected, it is a real key on a hard, low limit — a runaway loop hits it and the next person gets nothing. Use it when a live call is the only way to check what you are working on, and:
ui_state hint start prints already
preselects it in the model dropdown, so running as-is costs you nothing and
picks the cheap model. Change it only when the model itself is what you are
testing.pytest --runpaid / --runprerelease read
OPENROUTER_API_KEY, which is deliberately a different variable).references/seeded_project.md has the details.
Never trust the first frame. A screenshot taken right after open or goto is
very often blank white, and nothing says so. snapshot is not a usable gate — the
DOM is complete before Chromium paints. Settle the page first:
playwright-cli run-code "async page => await page.waitForLoadState('networkidle')"
playwright-cli screenshot --filename=/tmp/ui.png
Never redirect a playwright-cli failure into /dev/null. It reports failures
on stdout and exits 1, so >/dev/null 2>&1 — the thing you add to keep a
transcript readable — hides the entire reason a command failed. Discard output if
you must, but then test $?.
Never guess a locator's role. Kiln styles links as buttons. Run find first and
write the locator from what it reports — and prefer clicking the ref it gives you,
which is a single element by construction and cannot go strict-mode ambiguous.
Load only what the task needs.
| File | When |
|---|---|
| references/driving_the_ui.md | The command surface, and the locator traps in Kiln's UI specifically — duplicate labels, dropdowns, collapsed sections |
| references/e2e_suite.md | Running, reading, or debugging npm run tests:e2e |
| references/seeded_project.md | What the fixture contains, screen by screen, and whether a model provider is connected. Check here before concluding a screen is broken because it came up empty |
| references/search_tool.md | The Docs & Search / RAG part of the fixture, and what a sandbox with no API key can still do with it |
| references/extending_the_fixture.md | reset and snapshot — changing what future sessions start from |
还没有评价。欢迎分享你的使用感受。
无需登录即可发表。欢迎分享使用感受、提问，或说明在哪些环境下无法运行。
还没有评论。来写第一条吧。
Add AI models to Kiln or find available models missing from its list. The skill checks provider model identifiers and capabilities, updates model settings, runs tests against real APIs, and drafts a Discord announcement.
中文简介正在准备中。当前显示英语简介。
Find documentation and agent instructions made outdated by a branch’s code changes. Update affected paths, commands, settings, and behavior descriptions in place, then commit the documentation fixes alongside the code in the same pull request.
中文简介正在准备中。当前显示英语简介。
Draft first replies to customers and identify tickets that require human review before sending. The skill defines sensitive cases to escalate and guides replies that explain the next action, name who handles it, and match the customer's tone.
中文简介正在准备中。当前显示英语简介。
Audit Kiln’s model list against provider APIs and catalogs to identify removed models and upcoming retirements. Get findings organized by provider, then mark affected entries as deprecated after reviewing and approving the proposed changes.
中文简介正在准备中。当前显示英语简介。
Check whether Kiln’s base models still support fine-tuning, the process of adapting a model with additional training. Compare configured identifiers and Fireworks’ allowlist with provider information, then report unsupported entries and recommend specific updates.
中文简介正在准备中。当前显示英语简介。
Apply Kiln’s coding rules when editing or reviewing Python, TypeScript, and Svelte. The skill combines area-specific guidance with a script that checks changed lines, plus manual checks for architecture, configuration handling, shared state, and asynchronous operations.
中文简介正在准备中。当前显示英语简介。