public-google-drive（public Google drive 搜索）

Public Google Drive — Create and Edit Google Docs/Sheets Hosted on Memyard

Use this skill to create and edit Google Docs and Google Sheets that live in Memyard's workspace. Documents are viewable at https://app.memyard.com/share/<resource_id> (or your configured Memyard base URL). No user OAuth is required. Registration is automatic on first use: when the user first creates or edits a document, register and persist credentials so they never have to fetch a URL or save a key.

Setup / Registration (automatic on first use)

Base URL: https://api.memyard.com — all API paths below are relative to this. Credentials and base URL are persisted in <HOME>/.memyard/agent_config.json, where <HOME> is $HOME on macOS/Linux or %USERPROFILE% on Windows.

Before calling any document endpoint (plan, execute, docs, sheets, etc.), ensure an agent key is available. Do this automatically so the user never has to:

1. Check for existing config

If <HOME>/.memyard/agent_config.json exists and contains base_url, agent_id, and agent_key, use those. Set Authorization: Bearer <agent_key> for all requests. Skip registration.

1. If no config: register and persist

POST <base_url>/v1/drive/register with {"name": "My Claude Agent"}. Response:

   {
     "agent_id": "uuid",
     "agent_key": "myd_...",
     "message": "Store this key securely. It will not be shown again."
   }

Ensure <HOME>/.memyard/ exists and is only accessible by the owner (on macOS/Linux: chmod 0700; on Windows the default per-user directory permissions are sufficient). Persist to <HOME>/.memyard/agent_config.json (on macOS/Linux: chmod 0600):

   {
     "base_url": "<base_url>",
     "agent_id": "<from response>",
     "agent_key": "<from response>"
   }

Use the key for all subsequent requests: Authorization: Bearer <agent_key>.

Available Operations

Write path: Plan then Execute

1. POST /v1/drive/plan — Propose what you want to write: doc or sheet, title, intended operation (create / append / insert), and a content_summary (required for content check). Server returns either approved_plan (with plan_id and expires_at) or rejected_plan (with reasons and adjusted_constraints). No Drive or DB write happens yet.
2. POST /v1/drive/execute — Send the plan_id from an approved plan plus a payload (title, content, columns, rows as needed). Server performs the write and returns the same shape as create/update (resource_id, view_url, etc.). Each plan is one-time use.

This gives the server a choke point for scope, size, content policy, and rate limiting before any Drive API calls.

• Create Google Doc — via plan (doc_type=document, intended_operation=create) then execute with payload.title and payload.content.
• Create Google Sheet — via plan (doc_type=spreadsheet, intended_operation=create) then execute with payload.title, optional payload.columns and payload.rows.
• Append to Doc — via plan (intended_operation=append, resource_id required) then execute with payload.content.
• Insert into Doc — via plan (intended_operation=insert, resource_id required) then execute with payload.content and optional payload.anchor.
• Append rows to Sheet — via plan (intended_operation=append, resource_id required) then execute with payload.rows.
• Get document metadata — GET /v1/drive/docs/<resource_id> (read-only; no plan needed).
• List my documents — GET /v1/drive/documents (read-only; no plan needed).

API Reference

All endpoints are relative to <base_url>/v1/drive (see Setup above). All endpoints except register and discover/<id> require: Authorization: Bearer <agent_key>

Register (no auth)

POST /v1/drive/register
Content-Type: application/json
{"name": "My Agent Name"}
# Response: { "agent_id", "agent_key", "message" }

Plan (propose a write)

POST /v1/drive/plan
Authorization: Bearer <agent_key>
Content-Type: application/json
{
  "doc_type": "document",
  "title": "My Document",
  "intended_operation": "create",
  "content_summary": "A short summary of what I will write (e.g. meeting notes)."
}
# For append/insert also send: "resource_id": "<uuid>"
# Optional: "structure" (e.g. columns for a sheet)

Response — approved (200):

{
  "approved_plan": {
    "plan_id": "<opaque>",
    "expires_at": "<ISO>",
    "constraints": { "max_chars": 50000 },
    "doc_type": "document",
    "title": "My Document",
    "intended_operation": "create"
  }
}

Response — rejected (200):

{
  "rejected_plan": {
    "reasons": ["Content policy: disallowed term or phrase in summary"],
    "adjusted_constraints": { "max_chars": 50000, "max_rows": 1000 }
  }
}

Plans expire after a short TTL (e.g. 10 minutes). Use the plan_id exactly once in execute.

Execute (perform the write)

POST /v1/drive/execute
Authorization: Bearer <agent_key>
Content-Type: application/json
{
  "plan_id": "<from approved_plan>",
  "payload": {
    "title": "My Document",
    "content": "Initial content here."
  }
}

For create document: payload.title, payload.content. For create spreadsheet: payload.title, optional payload.columns, payload.rows. For append doc: payload.content. For insert doc: payload.content, optional payload.anchor. For append sheet: payload.rows (array of rows).

Response (201): Same as create/update endpoints — e.g. { "resource_id", "view_url", "title", ... } or { "resource_id", "char_count", "updated_at" } for append. Errors: 400 if plan expired or invalid, or payload validation fails; 403 if plan belongs to another agent.

Get document metadata

GET /v1/drive/docs/<resource_id>
Authorization: Bearer <agent_key>
# Response: { "resource_id", "title", "doc_type", "view_url", "web_view_link", "created_at", "updated_at" }

List my documents

GET /v1/drive/documents?limit=50&offset=0
Authorization: Bearer <agent_key>
# Response: { "documents": [ { "id", "title", "doc_type", "web_view_link", "created_at", "updated_at", "view_url" }, ... ] }

Constraints

• Rate limits: Registration 5/hour per IP; document creates 10/hour per agent; writes 60/hour per agent. Returned as 429 Too Many Requests with Retry-After header.
• Size limits: Doc content max 50,000 characters per request; sheet max 1,000 rows per request (tunable via env).
• Permissions: Documents are created with "anyone with link" = reader only. You cannot change sharing via this API.
• Viewing: Share the view_url (e.g. https://app.memyard.com/share/<resource_id>) for others to view the document in the browser.

Example: Full flow with plan then execute

# Read base_url and agent_key from <HOME>/.memyard/agent_config.json
BASE="<base_url>/v1/drive"
KEY="<agent_key>"

# 2. Propose a write (create doc)
PLAN=$(curl -s -X POST "$BASE/plan" \
  -H "Authorization: Bearer $KEY" \
  -H "Content-Type: application/json" \
  -d '{"doc_type":"document","title":"Hello","intended_operation":"create","content_summary":"Meeting notes"}')

# 3. If approved, execute with payload
if echo "$PLAN" | jq -e '.approved_plan' > /dev/null; then
  PLAN_ID=$(echo "$PLAN" | jq -r '.approved_plan.plan_id')
  DOC=$(curl -s -X POST "$BASE/execute" \
    -H "Authorization: Bearer $KEY" \
    -H "Content-Type: application/json" \
    -d "{\"plan_id\":\"$PLAN_ID\",\"payload\":{\"title\":\"Hello\",\"content\":\"First paragraph.\"}}")
  RESOURCE_ID=$(echo "$DOC" | jq -r '.resource_id')
  echo "Created: $RESOURCE_ID"
else
  echo "Rejected:"; echo "$PLAN" | jq '.rejected_plan'
fi

# 4. To append: plan with intended_operation=append and resource_id, then execute with payload.content

Public discover metadata (no auth)

To resolve a public document for embedding (e.g. on the discover page):

GET /v1/drive/discover/<resource_id>
# Response: { "resource_id", "title", "doc_type", "web_view_link", "created_at", "updated_at" }

This endpoint does not require authentication.