All guides › Skills and tools
Skills and toolsSkill catalog: the kaloko skill and the tools we recommend next to it
What the kaloko skill makes an agent do, what each reference covers, and which third-party skills and tools fit alongside, each with its implications: what runs where, which data leaves the machine, when not to use it.
Without a skill, an agent guesses when to walk a flow, what counts as done and whether it may touch production. A skill is a set of rules it follows without being reminded. Every skill you add changes what your agent does on its own, which commands it runs and which data leaves the machine, and most catalogs list skills without saying so. This one says it for the kaloko skill and for each tool we recommend next to it.
The kaloko skill
The router SKILL.md states when to use it, what done means, what is allowed without asking (local and review environments, test data there, fixing scenario mechanics) and what needs a person (changing criteria, production, uploading). Details live in references:
- First run (
reference/first-run.md): install,kaloko login, the sample run in a new organization, the first flow from a URL or a task. Implication: the agent signs in through the browser only once you approve the code. - Acceptance (
reference/acceptance.md): writingACCEPTANCE.mdandscenario.yml; one check per criterion; deterministic for countable things, semantic for judgement, manual for a reviewer; check packs (seo, a11y, perf and more) as advisory health, the design pack as acceptance; consent dialogs as the default first step. Implication: the plan is the source of truth, the agent never widens scope on its own. - Execution (
reference/execution.md): start, walk, capture, mail, the shared browser per environment, walking only what a change reaches (--affected), several locales at once and selectors that moved (--heal). Implication: one Chrome profile per environment; a production run never shares cookies with local work. - Evaluation (
reference/evaluation.md): how deterministic results and the semantic evaluator combine, the vision evaluator and its one-sentence reasons, thresholds, reporting without inventing. Implication: the text evaluator sees a reduced outline; the vision evaluator sees a scaled screenshot, only forvision: truecriteria and unsure answers. - Environments (
reference/environments.md): what may be done where, gates, accounts and saved sessions; production always read-only. Implication: the CLI blocks writes on production even if the agent tries. - Browsers and devices (
reference/devices.md): WebKit and Firefox, phone, tablet and TV presets, dark mode, reduced motion, installed web apps. Implication: every combination is a capture, so the agent keeps the matrix to what the task is about. - Apps (
reference/apps.md): Android over adb, the iOS Simulator and real iPhones, Electron, macOS and Windows apps, or screens from Maestro, Appium or XCUITest with their UI tree. Implication: emulators and simulators are the agent's to reset; real devices and real accounts need a person. - Stacks (
reference/stacks.md): one step for many pages of one template (articles, products, an SEO catalogue) from a sitemap, a listing, a file or a script. Implication: deterministic checks on every page, semantic ones on a sample, so the cost stays bounded. - Import (
reference/import.md): screenshots, Playwright reports, traces and agent walkthroughs laid out as a run. Implication: nothing is walked again; what you import is what reviewers see. - Recording (
reference/recording.md): video of a step or a whole flow, traces and the walkthrough video. Implication: recordings take storage, so the agent records when a screenshot cannot show the problem. - CI (
reference/ci.md): a workflow that walks each pull request's preview and comments on it. Implication: the agent asks before it writes or replaces a workflow. - Collaboration (
reference/collaboration.md): share, PR comment, review mode and approvals carried over unchanged steps, feedback, inbox, verdicts, baselines, compare, the signed acceptance record, clean-up. Implication: the agent reads feedback before it continues and never approves for a person. - Integrations (
reference/integrations.md): issues in Jira and Linear, channels per project, hosted runs on a schedule. Implication: one issue per real problem; channels and connections are an admin's job. - MCP (
reference/mcp.md): runs, feedback, verdicts, the inbox, comments, the review queue and issues as tools. Implication: writes need a token with the right scope; capturing and uploading stay in the CLI. - Docs (
reference/docs.md): the visual changelog and guides with step references, stale sections, translations, exports, release tools. Implication: the agent drafts and checks; recording a release and publishing docs wait for a person. - Documentation (
reference/documentation.md): manuals and courses refreshed from the newest accepted captures. Implication: only accepted runs feed documentation when you ask for it. - Design (
reference/design.md,reference/portable-html.md): live HTML steps, drafts and versions, Figma import, the screen library, the step spec and drift for implementing an approved design. Implication: a version is published and a reference set only when a person says so.
Recommended alongside
- Playwright MCP (browser control for the agent): connect it to the CDP endpoint of
kaloko browser startso agent-driven steps happen in the same profile as scripted ones. Implication: the agent sees page content locally; nothing goes to a third party. - axe (accessibility rules): the
a11ypack covers names, labels, headings, contrast, landmarks and the Tab walk; for axe's full rule set run it in a step script and store the result as an attachment. Implication: local only. - Lighthouse (performance): the
perfpack already reads LCP, CLS, TTFB and page weight from the walk; for a full report use the same pattern, one report per key page. Implication: noisy on shared runners; compare trends, not single numbers. - Test-data seeding in your project: a
preparestep for review and staging so scenarios create only what they need. Implication: never on production.
Where the data goes
- Screenshots, HTML, e-mails: your machine and your Kaloko organization (Cloudflare storage, EU-law operator). Files are served from a separate origin under short-lived signed links.
- Semantic evaluation: a reduced text outline with e-mails and tokens masked to the evaluator you configured (default JEV by TypeSafe), only for
semanticcriteria. The vision evaluator (Anthropic, with your key or through Kaloko) gets a scaled screenshot, only forvision: truecriteria and answers the text evaluator is unsure about;evaluators: { vision: false }turns it off. - Nothing to Sinfin beyond what the service needs to show your canvas.
FAQ
Can I write my own sub-skill?
Yes: project skills may add conventions (paths, test data, locales) and the kaloko skill follows them where they are more specific.
Does the skill work with agents other than Claude Code and Codex?
Any agent that can run shell commands can follow SKILL.md; kaloko init without --agent installs nothing and leaves the CLI for scripts and CI.