Menu
How the TriageAgent should approach a GitHub issue. Load this before deciding whether to attempt a fix or to write up findings.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Describes how to write pull/merge requests. Use when asked to write or edit a pull request or merge request description. This skill is not relevant to commit messages.
Index of Cloudflare-platform skills relevant to this repo. Load when working on anything that touches Workers, Durable Objects, wrangler, the sandbox SDK, the agents SDK, or other Cloudflare platform surface. Triggers include "Cloudflare", "Workers", "Durable Object", "wrangler", "KV", "R2", "D1", "Vectorize", "Containers", "sandbox".
Debug wsd in real-FUSE mode end-to-end without workerd, vitest-pool-workers, or wrangler in the loop. Boot the linux-x64 binary in a privileged docker container, drive its capnweb /ws endpoint from Node, simulate DO-side sync from a SQLiteTestStorage, and isolate FUSE-related deadlocks. Load when a real-FUSE bug reproduces locally but unit tests pass, when the harness vitest tests hang against a real container, or when you need to attribute a wedge to FUSE vs sync vs exec.
capnweb RPC patterns for this repo, with a heavy focus on stub lifecycle and disposal. Load when touching anything that crosses the Durable Object ↔ wsd boundary: packages/rpc, packages/workspace, the wsd client, or the Durable Object server. Triggers include "capnweb", "RpcTarget", "stub", "RPC wire", "promise pipelining", "stub disposal", "RpcPromise".
Style guidance for the written word. Use when asked to write code comments, commit messages, pull requests, READMEs, or documentation. Triggers might be phrases like "write a commit", "document this", or "add a README".
Drives development with tests. Use when implementing any logic, fixing any bug, or changing any behavior. Use when you need to prove that code works, when a bug report arrives, or when you're about to modify existing functionality.
| name | triage |
| description | How the TriageAgent should approach a GitHub issue. Load this before deciding whether to attempt a fix or to write up findings. |
This skill captures the playbook the TriageAgent should follow when
handed a GitHub issue. It is mounted into the workspace from an R2
bucket at /workspace/.agents/skills/triage/SKILL.md; the agent can
read it like any other file.
/workspace/repo with git_clone.
Always shallow (depth: 1) unless the issue specifically asks
about history.read and ls rather
than exec cat / exec ls. Skim the README and any file paths
the issue mentions before going wider.write /
edit, then run the project's own tests / typecheck / build via
exec to confirm. Stop as soon as the verification passes.report_update once or twice. Keep
messages terse (one or two sentences). Do not say "DONE"
yourself; the workflow emits the terminal message.When the turn is finished, reply with a short natural-language summary covering:
fix: …, docs: …,
investigate: …)The workflow computes the unified diff itself via git diff HEAD;
do not include it.
Reach for the dedicated tools before exec. They are faster, give
structured output, and don't depend on which binaries happen to live
in the container:
| Need | Use |
|---|---|
| Clone a repo | git_clone |
| Read a file | read |
| List a directory | ls |
| Modify a file | write or edit |
| Run project tests / typecheck / build | exec |
| Update the user | report_update |
exec is the escape hatch — use it only for things the dedicated
tools can't express.