One search for your own team's files and the public ecosystem. See how real projects write their agent instructions, and bring the good ideas home.
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
CloudflareSkillcloudflare/workerd/.opencode/skills/update-v8/SKILL.md
Step-by-step guide for updating the V8 JavaScript engine in workerd, including patch rebasing, dependency updates, integrity hashes, and verification. Load this skill when performing or assisting with a V8 version bump.
CloudflareSkillcloudflare/workerd/.opencode/skills/wd-test-format/SKILL.md
Detailed guide for authoring .wd-test files in workerd, with examples of bindings, Durable Objects, multi-service configs, TypeScript tests, and network access.
CloudflareSkillcloudflare/workerd/.opencode/skills/workerd-api-review/SKILL.md
Performance optimization, API design & compatibility, security vulnerabilities, and standards spec compliance for workerd code review. Covers tcmalloc-aware perf analysis, compat flags, autogates, web standards adherence, and security patterns. Load this skill when reviewing API changes, performance-sensitive code, security-relevant code, or standards implementations.
CloudflareSkillcloudflare/workerd/.opencode/skills/workerd-safety-review/SKILL.md
Memory safety, thread safety, concurrency, and critical detection patterns for workerd code review. Covers V8/KJ boundary hazards, lifetime management, cross-thread safety, and coroutine pitfalls. Load this skill when reviewing any C++ code.
CloudflareSkillcloudflare/workerd/.opencode/skills/wpt-update/SKILL.md
Updates and triages the Web Platform Tests dependency in workerd. Use when asked to bump, update, or triage WPT.
CloudflareAGENTS.mdcloudflare/workerd/src/bench/streams/AGENTS.md
Comparative benchmarks of the two streams implementations, the legacy C++ one (src/workerd/api/streams/) and the TypeScript one (src/per_isolate/webstreams/), organized like src/tests/streams/: one subdirectory per functional area. Every benchmark runs the same code under both implementations, so the two reports compare case by case. Correctness and parity are src/tests/streams/'s job; these suites only measure. - Each named export of main.js is an entrypoint with a bench() handler, which registers a group of cases with b.run(). The report names a group <service>:<export> (streams-readable-ts:readLoop).…
CloudflareAGENTS.mdcloudflare/workerd/src/bench/streams/body/AGENTS.md
Where the runtime's bodies meet streams, under both implementations: the runtime reading JavaScript streams as bodies, JavaScript reading bodies the runtime produces, and the APIs built on bodies. Suite rules: src/bench/streams/AGENTS.md. The JavaScript stream shapes are 256 x 64 B, 64 x 4 KiB and 16 x 64 KiB (SHAPES in helpers.js). Bodies and writes copy their bytes, so each size's buffer is shared. The configs bind SELF to the worker itself. main.js's default export is service-binding.js's fetch handler, which…
CloudflareAGENTS.mdcloudflare/workerd/src/bench/streams/compression/AGENTS.md
CompressionStream and DecompressionStream, under both implementations. Suite rules: src/bench/streams/AGENTS.md. The payload (PAYLOAD in helpers.js) is 64 KiB of markup-like text, which compresses to about 9 KiB. - Both implementations drive the same native zlib codec, so the small chunks isolate the cost of the streams around it. - Under the compression-rs autogate (@all-autogates), the codec is zlib-rs, whose output depends on how its input is split. The expected compressed sizes therefore come from compressing the same chunks once in bench(),…
CloudflareAGENTS.mdcloudflare/workerd/src/bench/streams/digest/AGENTS.md
crypto.DigestStream, under both implementations. Suite rules: src/bench/streams/AGENTS.md. The byte shapes are 256 x 64 B, 64 x 4 KiB and 16 x 64 KiB (SHAPES in helpers.js). Each case checks the digest's size and the stream's bytesWritten. - Both implementations drive the same native digest context, so the small chunks isolate the cost of the stream around it.
CloudflareAGENTS.mdcloudflare/workerd/src/bench/streams/encoding/AGENTS.md
TextEncoderStream and TextDecoderStream, under both implementations. Suite rules: src/bench/streams/AGENTS.md. The shapes are 256 x 16, 64 x 4 Ki and 16 x 64 Ki characters or bytes (SHAPES in helpers.js): a stream of short chunks, like generated tokens, and longer ones. TEXT names the text: ASCII, and two- and three-byte characters in UTF-8. - Both TextDecoderStreams decode with TextDecoder's native decoder, so the decoder differences are in the streams around it. - The TextEncoderStreams encode differently. The TypeScript one calls…
CloudflareAGENTS.mdcloudflare/workerd/src/bench/streams/identity/AGENTS.md
IdentityTransformStream and FixedLengthStream, under both implementations. Suite rules: src/bench/streams/AGENTS.md. The byte shapes are 256 x 64 B, 64 x 4 KiB and 16 x 64 KiB (SHAPES in helpers.js). Writes copy their bytes, so each size's chunk is one shared buffer. - The implementations split and merge writes into reads differently, so the cases count bytes, not reads. Over several queued writes, a BYOB read takes one write's bytes under C++ (stopping once a minimum is met), while the TypeScript…
CloudflareAGENTS.mdcloudflare/workerd/src/bench/streams/piping/AGENTS.md
pipeTo() and pipeThrough(), under both implementations. Suite rules: src/bench/streams/AGENTS.md. Pipes into identity streams are in identity/, and pipes of Response bodies are in body/.
CloudflareAGENTS.mdcloudflare/workerd/src/bench/streams/readable/AGENTS.md
ReadableStream with the default controller, under both implementations. Suite rules: src/bench/streams/AGENTS.md. - A tee branch's cancel() settles only once the source finishes or the other branch cancels too under the TypeScript implementation, as in the spec; C++ settles it at once. read one branch, cancel the other awaits the cancel together with the other branch's read, which both implementations complete. - Once a source has called close(), C++ cancels the stream without running the source's cancel(), even while chunks are…
CloudflareAGENTS.mdcloudflare/workerd/src/bench/streams/readable-byte/AGENTS.md
ReadableStream with type: 'bytes', under both implementations. Suite rules: src/bench/streams/AGENTS.md. The byte shapes are 256 x 64 B, 64 x 4 KiB and 16 x 64 KiB (SHAPES in helpers.js): many small chunks, a common network read size, and large chunks. - For a default read of a byte stream without autoAllocateChunkSize, C++ allocates a 4 KiB BYOB request (16 KiB under an autogate) where the TypeScript implementation, like the spec, has none (readable-byte test ledger #5). defaultReader's pull() source…
CloudflareAGENTS.mdcloudflare/workerd/src/bench/streams/strategies/AGENTS.md
CountQueuingStrategy and ByteLengthQueuingStrategy, under both implementations. Suite rules: src/bench/streams/AGENTS.md. - The strategy-object cases are the suites' only synchronous ones: they queue no microtasks. Every result goes through b.blackBox(), whose cost per call the report shows in the group's header, the same in both cells. - Under C++, every access to a strategy's size mints a new function, where the TypeScript implementation returns one shared function (strategies test ledger #1). The size() cases, which read size for each call, therefore also…
CloudflareAGENTS.mdcloudflare/workerd/src/bench/streams/transform/AGENTS.md
TransformStream, under both implementations. Suite rules: src/bench/streams/AGENTS.md. TRANSFORMERS in helpers.js names the transformers the cases compare: none, transform() enqueuing each chunk, an async transform(), one enqueuing each chunk twice, one dropping every other chunk, and flush() enqueuing a last chunk. Each stream gets a new transformer.
CloudflareAGENTS.mdcloudflare/workerd/src/bench/streams/writable/AGENTS.md
WritableStream and its default writer, under both implementations. Suite rules: src/bench/streams/AGENTS.md.
CloudflareAGENTS.mdcloudflare/workerd/src/cloudflare/AGENTS.md
TypeScript implementations of Cloudflare product APIs (AI, D1, R2, Vectorize, etc.). It is common, but not required, for top-level .ts files to re-export from internal/ via cloudflare-internal: specifiers. This allows for a clean separation between public API surface and internal implementation details. Each product test directory contains: Mock wiring uses wrapped bindings: moduleName = "cloudflare-internal:<product>-api" with innerBindings pointing fetcher at the mock service. Shared instrumentation-test-helper.js lives in internal/test/.
CloudflareAGENTS.mdcloudflare/workerd/src/node/AGENTS.md
TypeScript and JavaScript layer implementing Node.js compatible built-in modules for Workers. It is split across multiple layers: It is common, but not required, for top-level .ts files to re-export from internal/ via node-internal: specifiers. This allows for a clean separation between public API surface and internal implementation details. See README.md for 12 policy rules governing compat scope and philosophy. node:* modules are gated behind the nodejs_compat compatibility flag. The node:async_hooks module can be enabled individually via the nodejs_als compatibility flag.
CloudflareAGENTS.mdcloudflare/workerd/src/per_isolate/AGENTS.md
Per-isolate JavaScript/TypeScript bootstrap: scripts that run synchronously at context creation, before any user code. C++ entry point: src/workerd/api/per-isolate-bootstrap.c++. - Scripts are compiled as functions whose parameters are the pseudo-globals require, module, exports, compatFlags, autogates, primordials, utils are in scope but NOT on globalThis (see per_isolate-env.d.ts). - Module system is bootstrap CommonJS: require('webstreams/queue') + module.exports = {...}. The src/node/ ESM-only rule does NOT apply here. Circular requires are a FATAL startup error — keep modules acyclic (e.g., webstreams/ring-buffer is a leaf,…
Plain text files in a repository that tell a coding agent how the project works: commands to run, conventions to follow and things to avoid. CLAUDE.md, AGENTS.md, cursor rules and skills are the common kinds.
CLAUDE.md is read by Claude Code. AGENTS.md is an open format that Codex, Cursor and other agents read. Many projects keep one and point the other at it.
A folder with a SKILL.md that describes one capability, such as filling PDFs or reviewing code. The agent loads it only when the task calls for it.
Your agents already can, over MCP, limited to the files you're allowed to read. Searching them from this page is coming.
