Cloudflare R2 Upload

Cloudflare R2 Upload

Use this skill when the task requires pushing local files to a Cloudflare R2 bucket and producing stable public URLs through a bound custom domain (for example object.innei.in).

Scope

• Target task class: single-file or batched uploads into one R2 bucket under a specified key prefix.
• Supported auth paths:
  • wrangler CLI with an existing OAuth session.
  • wrangler CLI with CLOUDFLARE_API_TOKEN env (from a Cloudflare API token, e.g. stored as R2_API_KEY).
• Out of scope: bucket creation, custom-domain binding, CORS / lifecycle config, signed URLs, multipart uploads of files >300 MB.

Inputs

Workflow

textCopy code
[1] Identify account + bucket
      -> wrangler whoami to list accounts
      -> wrangler r2 bucket list (per candidate account) to find BUCKET

[2] Upload each file
      -> wrangler r2 object put BUCKET/KEY --file=PATH --content-type=MIME --remote

[3] Verify via public domain
      -> curl -sI https://PUBLIC_DOMAIN/KEY and assert HTTP 200

[4] Report final URLs
      -> https://PUBLIC_DOMAIN/KEY for each uploaded file

Auth Setup

Prefer an existing wrangler OAuth session — simplest path:

bashCopy code
wrangler whoami

If OAuth is unavailable, use an API token instead:

bashCopy code
export CLOUDFLARE_API_TOKEN="$R2_API_KEY"

If the user belongs to multiple accounts, always set CLOUDFLARE_ACCOUNT_ID explicitly; otherwise bucket list / object put may default to the wrong account and fail with code: 10042 — Please enable R2 through the Cloudflare Dashboard, which actually means "this account has no R2, try another account".

Finding the Right Account

wrangler whoami prints an account table. For each account id, list buckets until the target bucket is found:

bashCopy code
for acct in <id1> <id2> <id3>; do
  echo "=== $acct ==="
  CLOUDFLARE_ACCOUNT_ID="$acct" wrangler r2 bucket list 2>&1 | grep -E "^name:|ERROR" | head -20
done

Cache the resolved CLOUDFLARE_ACCOUNT_ID for the remaining commands.

Single-File Upload

bashCopy code
CLOUDFLARE_ACCOUNT_ID="$ACCOUNT_ID" \
  wrangler r2 object put "$BUCKET/${KEY_PREFIX}<name>.webp" \
    --file="/path/to/local.webp" \
    --content-type="image/webp" \
    --remote

Flags that matter:

• --remote — required. Without it, the write goes to the local simulator, not production.
• --content-type — set explicitly. Default guessing is unreliable and bad MIME breaks CDN / browsers.
• Target is a single positional BUCKET/KEY string (no space between bucket and key).

Batch Upload (Directory)

bashCopy code
cd "$SRC_DIR" && for f in *.webp; do
  echo "--- $f ---"
  CLOUDFLARE_ACCOUNT_ID="$ACCOUNT_ID" \
    wrangler r2 object put "$BUCKET/${KEY_PREFIX}${f}" \
      --file="$f" \
      --content-type="image/webp" \
      --remote 2>&1 | grep -E 'Creating|error|ERROR' | head -3
done

For mixed MIME types, resolve per-file:

bashCopy code
mime_of() {
  case "$1" in
    *.webp) echo "image/webp" ;;
    *.png)  echo "image/png"  ;;
    *.jpg|*.jpeg) echo "image/jpeg" ;;
    *.svg)  echo "image/svg+xml" ;;
    *.json) echo "application/json" ;;
    *.html) echo "text/html; charset=utf-8" ;;
    *.txt|*.md) echo "text/plain; charset=utf-8" ;;
    *)      echo "application/octet-stream" ;;
  esac
}

Verification

Never claim success based on wrangler output alone. Run an HTTP HEAD against the public domain for every uploaded key:

bashCopy code
for f in "$SRC_DIR"/*.webp; do
  name=$(basename "$f")
  code=$(curl -sI -o /dev/null -w '%{http_code}' \
    "https://${PUBLIC_DOMAIN}/${KEY_PREFIX}${name}")
  printf "%-40s %s\n" "$name" "$code"
done

All lines must show 200. A 404 means the key path or bucket binding is wrong; a 403 means the custom domain is not bound to this bucket.

Common Mistakes

Rules

• Prefer OAuth session over API token when both are available.
• Always verify at least one URL per batch through the public domain before reporting completion.
• Do not delete previous-generation objects immediately after updating DB records; keep a rollback window.
• Never paste R2_API_KEY or CLOUDFLARE_API_TOKEN values into commit logs, commit messages, or PR descriptions.
• Do not mix --remote and non-remote calls in the same batch.

Reporting Output

Preferred structured summary after a batch:

Always list the fully-qualified public URLs at the end so the caller can paste them directly into downstream config or DB updates.