Setting up aicorn

Setting up aicorn

Write a per-project configuration file at .claude/aicorn.local.md that the companion fetch skill reads to decide where and how to route WebFetch calls. The file uses YAML frontmatter for machine-readable values.

Inputs

Three values are needed:

1. worker_url — base URL of an agentify Worker (e.g. https://agentify.mikhailnovikov.workers.dev). No trailing slash, no /fetch suffix.
2. username — sent as the user query parameter on every routed fetch. Any string the user prefers.
3. domains — comma-separated allowlist of hostnames ("wikipedia.org,auchan.pt") OR the literal string * for all hosts.

Step 1 — Collect values

Positional arguments are available as $1 (worker_url), $2 (username), $3 (domains). An empty or missing argument means "ask the user".

Before prompting, attempt to read any existing .claude/aicorn.local.md (use the Read tool). If it exists, use its current frontmatter values as the defaults shown to the user — that way running /aicorn:setup to change just one field doesn't blow away the others.

When values are missing, ask via AskUserQuestion with these defaults:

• worker_url default: existing value if present, else https://agentify.mikhailnovikov.workers.dev.
• username default: existing value if present, else leave blank — the user must type one.
• domains default: existing value if present, else *.

Step 2 — Validate and normalise

Apply each rule. On failure, auto-fix where safe and confirm with the user before proceeding; reject and re-prompt only when no safe fix exists.

Field	Rule	If broken
worker_url	Must start with http:// or https://.	Reject and re-prompt.
worker_url	Trailing /.	Strip silently.
worker_url	Contains /fetch or other path.	Reject and re-prompt — the user supplied the wrong URL.
username	Non-empty.	Reject and re-prompt.
username	Contains whitespace, ?, &, /, #.	Replace each bad char with -, show the cleaned value, ask the user to confirm or supply a different one.
domains	Equals * OR comma-separated tokens each matching ^[a-z0-9.-]+$.	Reject and re-prompt — common mistakes are leaving in https:// or trailing paths.

Step 3 — Health-check the Worker

Before writing the config, verify the Worker is reachable. Substitute the validated worker_url directly into the curl command — do NOT rely on environment variables. Run via Bash:

curl -fsS -m 5 "https://agentify.mikhailnovikov.workers.dev/" 2>&1

(replace the URL with the actual worker_url value at runtime).

Expected response body: agentify ok.

• Reachable: continue.
• Non-200 / timeout / network error: tell the user the Worker is unreachable, ask whether to save anyway (perhaps the Worker has not been deployed yet) or abort. Default to abort.

Step 4 — Write .claude/aicorn.local.md

Create the .claude directory if it does not exist (mkdir -p .claude). Write the file via the Write tool, OVERWRITING any prior content. The defaults pulled in Step 1 ensure no field gets reset by accident.

File content — substitute the validated values for the placeholders shown in <>. Do NOT copy the placeholder names verbatim into the file:

---
worker_url: <WORKER_URL>
username: <USERNAME>
domains: <DOMAINS>
---

# aicorn local config

Edit the YAML frontmatter above to change settings. The `fetch` skill
reads these values on the first WebFetch of each session.

- `worker_url` — base URL of the agentify Worker. No trailing slash, no /fetch.
- `username` — sent as `?user=...` on every /fetch call.
- `domains` — comma-separated hostname allowlist, or `*` for all hosts.

YAML gotcha: if domains is the literal *, write it quoted as domains: "*". Bare * is interpreted as a YAML alias and will fail to parse.

Step 5 — Add to .gitignore

Run from the project's git root so the entry lands in the right .gitignore:

cd "$(git rev-parse --show-toplevel 2>/dev/null || pwd)"
grep -qxF '.claude/aicorn.local.md' .gitignore 2>/dev/null \
  || echo '.claude/aicorn.local.md' >> .gitignore

If .gitignore does not exist, do NOT create one — the user may have intentional reasons. Just skip and tell them: "No .gitignore present — the config file at .claude/aicorn.local.md may end up in git history. Consider adding it manually if that matters."

A redundant entry is possible if the user already ignores .claude/ more broadly — that is harmless.

Step 6 — Confirm

Print:

Saved .claude/aicorn.local.md
  worker_url: <url>
  username:   <name>
  domains:    <list>

The fetch skill will route WebFetch calls through this Worker.
Try `fetch https://en.wikipedia.org/wiki/Cache` to verify routing.

If the health check failed and the user opted to save anyway, append: "Note: Worker was unreachable at save time; routed fetches will fail until the Worker is online."

Re-running setup

This skill always overwrites the existing file. Step 1's "read existing values as defaults" preserves any field the user does not change. To wipe the config entirely, the user can rm .claude/aicorn.local.md first.