cloudflare/workerd
21 skills778 installs177k starsGitHub
Install
npx skills add https://github.com/cloudflare/workerdSkills in this repo
1Test Driven InvestigationThe test-driven-investigation skill reframes bug work around writing a failing reproducer as early as possible rather than spiraling through repeated source reads. Its core principle is that a twenty-line failing test reveals more than hours of mental modeling. The workflow reads the error or assertion, identifies the minimal trigger state and operation, writes a test, runs it, and lets results guide the next step within bounded orientation to find APIs and test harnesses. Commit history checks with git log on relevant paths can narrow suspects but must not replace testing, and adjacent commits should be reviewed to avoid confirmation bias. Agents adapt existing subsystem tests instead of inventing from scratch, verify output proves the code path ran, and scope tests to the failure mechanism rather than full production stacks. Triggers include investigating crashes, assertions, or unexpected workerd behavior.49installs2Dad JokesThe dad-jokes skill instructs agents to deliver one safe-for-work joke after completing sustained work such as long builds, test suites, multi-step investigations, or large refactors. It fires roughly once every twenty to thirty tool calls of sustained effort, or after grueling debug sessions, but skips during production incidents or urgent fixes. Agents must load this skill before any user-requested joke and never improvise humor without it. Format selection rotates among pun, limerick, and Q&A dad joke using an incidental number seed or least-recently-used format to avoid repetitive Q&A patterns. Rules require one joke only, context-aware references to the specific API, bug, module, or concept just worked on, short length, no explanation, and no asking permission. Inspiration sources include Cap'n Proto and workerd runtime concepts, Bazel build quirks, and debugging motifs. Triggers include completing tasks over five tool calls or explicit user requests for humor.47installs3Markdown DraftsThe markdown-drafts skill ensures agent-drafted content intended for external systems preserves markdown syntax when users copy it elsewhere. It applies to GitHub issues and pull request descriptions, Jira tickets, Confluence or wiki pages, design documents, RFCs, Slack messages, and any content the user will paste outside chat. Formatting rules require hash headers, list syntax, fenced code blocks with language tags, inline backticks for identifiers and paths, bold and italic emphasis, markdown tables, blockquote callouts, linked references, horizontal rules, and task list checkboxes. Structure guidance favors a concise summary first, scannable sections, short paragraphs, and next steps at the end. Because chat UIs render markdown and strip raw syntax, the entire draft must be wrapped in a plain triple-backtick fence so users copy literal markup. Triggers include drafting issues, PR descriptions, Jira tickets, or wiki pages for external paste.46installs4Rust ReviewThe rust-review skill provides Rust code review guidance for Cloudflare workerd source under src/rust/. It covers CXX FFI safety, unsafe code patterns, JSG resource conventions, error handling, and a review checklist adapted from existing C++ review skills. Reviewers must always use the docs/reference/rust-review-checklist.md file when evaluating Rust changes rather than improvising criteria. The skill loads when reviewing Rust pull requests or diffs in the workerd repository where memory safety across the CXX boundary, JSG lifecycle rules, and workerd-specific error propagation matter. It complements C++ review skills by translating the same rigor to Rust modules that interface with the JavaScript runtime and V8 bindings. Use when a contributor submits Rust changes to workerd, when unsafe blocks need audit, or when FFI bindings between Rust and C++ require checklist-driven review before merge to production runtime code.42installs5Investigation NotesThe investigation-notes skill requires a lightweight scratch document as external memory during bug investigations in workerd. Create ~/tmp/investigate-short-name.md during orientation with sections for Error, Current Focus, Hypotheses, Code Read, Tests, Ruled Out, and Next. Rules cap active hypotheses at three with only one TESTING at a time, mandate updating after each test run before other work, and forbid large code dumps by referencing file:line instead. The document never takes priority over writing or running a test. Use when investigations span multiple files, hypotheses, or test iterations.41installs6Ts StyleThe ts-style skill points reviewers and authors to docs/reference/ts-style.md for JavaScript and TypeScript conventions in workerd. Load when reviewing or writing code under src/node/, src/cloudflare/, or JS/TS tests in src/workerd/. It covers TypeScript strictness, import and export patterns, private field syntax, error handling, feature gating, and test structure. Use during code review or authoring workerd TypeScript changes.40installs7Find And Run TestsThe find-and-run-tests skill explains locating and running tests in workerd and parent-project integration setups. workerd uses wd_test for JS/TS integration and kj_test for C++ unit tests with bazel query kind test patterns. Running prefers --test_output=streamed and --nocache_test_results. Parent projects may append variant suffixes like @all-autogates; cached results and filter flags that match zero tests are called out as hazards. Debugging steps include reading test.log, verifying feature evidence in output, and comparing subrequest headers. Load when unsure of exact bazel target names or invocation flags.39installs8Parent Project SkillsThe parent-project-skills skill bootstraps discovery of additional skills when workerd is embedded in a larger proprietary parent project. Check ../../.git or ../../AGENTS.md for a parent root, then read ../../.opencode/skills/*/SKILL.md for relevant guidance. Enforce a strict one-way boundary: internal context may inform reasoning but must never appear in workerd commits, comments, PR text, tests, or docs. Use for production Sentry investigations, parent integration tests, cross-repo debugging, and deployment-context API behavior without leaking internal details into open-source workerd artifacts.39installs9Workerd Api ReviewThe workerd-api-review skill points reviewers to docs/reference/api-review-checklist.md for performance optimization with tcmalloc awareness, API design and compatibility, security vulnerability patterns, compat flags, autogates, and web standards spec compliance. Load when reviewing API changes, performance-sensitive code, security-relevant paths, or standards implementations in workerd. Use during planning, writing, and reviewing workerd API surface changes.39installs10Workerd Safety ReviewThe workerd-safety-review skill directs reviewers to docs/reference/cpp-safety-review-checklist.md for memory safety, thread safety, concurrency correctness, V8 and KJ boundary hazards, lifetime management, and coroutine pitfalls. Load when reviewing any workerd C++ code changes. Use during planning, implementation, and review of performance-sensitive or concurrency-heavy workerd patches.39installs11Bazel Test HygieneThe bazel-test-hygiene skill enforces three rules before claiming test success in workerd. Always pass --nocache_test_results because stale binaries cause false positives. Never use --test_arg -f filters that silently pass with zero matches. Run the full //... suite before claiming a fix is complete and verify Executed N out of N tests all pass. Regression tests require red-green verification on the full suite: green with fix, red without fix, green again after restore. Anti-patterns table contrasts cached runs, filter misuse, and partial target claims.38installs12Update V8The update-v8 skill guides bumping the V8 JavaScript engine in workerd with human confirmation at each step. Prerequisites include depot_tools and a local V8 checkout outside the workerd repo. Steps identify target version from chromiumdash, sync old version, git am workerd patches onto workerd-patches branch, rebase onto new version with conflict review, format-patch regenerated patches, replace patches/v8 after human approval, update deps and integrity hashes, and run verification builds. Patch categories cover serialization, build system, embedder hooks, bug workarounds, API additions, and ICU config. Never drop patches or update hashes without explicit confirmation.37installs13Add AutogateThe add-autogate skill documents registering temporary AutogateKey enums in autogate.h and string mappings in autogate.c++ with kebab-case runtime names prefixed workerd-autogate-. Code guards use Autogate::isEnabled around new versus old paths. Testing uses @all-autogates wd_test variants, targeted C++ setup, or runtime config. Autogates differ from compat flags: temporary kill switches for internal rollout rather than permanent user-visible compatibility dates. Use when gradually enabling risky workerd internal behavior.36installs14Identify ReviewerThe identify-reviewer skill runs gh auth status and git config user.name and email in parallel to detect the local reviewer identity before PR or commit analysis. Matches GitHub handle, git author name, and email against PR comments and commits to attribute prior feedback in second person and note approval or changes-requested status. Distinguishes when the user is also PR author for self-review tone. If gh is unavailable, falls back to git config alone without blocking the review.36installs15Pr Review GuideThe pr-review-guide skill supports workerd pull request reviews by combining identify-reviewer for GitHub and git identity detection with commit-categories path-based changelog grouping. Reviewers load identify-reviewer first to attribute prior comments in second person and note approval status. commit-categories maps changed paths to API, Node compat, Python, Rust, Cloudflare APIs, I/O, JSG, Server, Build, Types, Docs, Tests, or Other with cross-cutting compat flag and autogate callouts. Use at the start of PR review, code review, or commit log analysis in workerd.36installs16Commit CategoriesThe commit-categories skill provides canonical path-based grouping for workerd commits into API, Node.js compat, Python, Rust, Cloudflare APIs, I/O, JSG, Server, Build, Types, Docs, Tests, or Other. Cross-cutting callouts highlight New or Updated Compat Flags and Autogates separately from primary categories. Categorization uses git diff-tree file lists and majority path matching. Must load before changelog or whats-new commit categorization commands. Use when building release notes or summarizing workerd commit batches.35installs17Kj StyleThe kj-style skill directs authors and reviewers to docs/reference/kj-style.md for KJ and workerd C++ conventions covering naming, types, memory management, error handling, inheritance, constness, and formatting. Load when reviewing or writing C++ in the workerd codebase. Complements workerd-safety-review and api-review checklists for style versus safety concerns.35installs18Wd Test FormatThe wd-test-format skill explains .wd-test files as Capn Proto workerd configs for JS/TS integration tests. Basic structure uses Workerd.Config services with embed modules, compatibilityFlags instead of compatibilityDate, and bindings for text, data, and service namespaces. Module types include esModule, json, wasm, and commonJsModule. Mandates loading reference/advanced-configs.md for Durable Objects, multi-service bindings, network access, and TypeScript tests. Use when writing or reviewing workerd integration test configs.35installs19Add Compat FlagThe add-compat-flag skill walks adding behavioral compatibility flags in workerd. Choose enable and optional disable snake_case flag names, add fields to CompatibilityFlags in compatibility-date.capnp using next-capnp-ordinal tool for ordinals, set compatEnableDate and annotations like experimental or neededByFl, access via FeatureFlags getters in C++, add wd_test coverage, and document in changelog categories. Flags allow breaking changes rolled out by compatibility date unlike temporary autogates. Use when introducing permanent user-visible behavioral toggles in workerd.34installs20Module RegistryThe module-registry skill documents workerd internal module registry mechanics for registering and resolving modules, used when implementing or debugging runtime module loading and build-time embedding.34installs21Verification Before Completionverification-before-completion is a Claude Code skill in the AI & Agent Building category.1installs