Agent Skills
› NeverSight/learn-skills.dev
› google-docs
google-docs
GitHub通过 Docs v1 REST API 读取、创建和编辑 Google Docs。支持获取文档文本、新建文档、插入或替换文本及导出内容，适用于涉及 Google 文档操作的任务。
Trigger Scenarios
用户提及 Google Doc
读取文档文本
创建文档
插入或替换文本
导出文档内容
Install
npx skills add NeverSight/learn-skills.dev --skill google-docs -g -y
SKILL.md
Frontmatter
{
"name": "google-docs",
"license": "Apache-2.0",
"metadata": {
"author": "acedatacloud",
"version": "1.0"
},
"connections": [
"google\/docs"
],
"description": "Read and edit Google Docs via the Docs v1 REST API. Use when the user mentions a Google Doc, reading a document's text, creating a doc, inserting or replacing text, or exporting a doc's content.",
"when_to_use": "Trigger when the user wants to read a Google Doc's content, create a\nnew doc, or insert \/ replace text in one. The connector grants\n`documents.readonly` by default; the user opts in to `documents`\n(read + write) at install — confirm before edits. Find the doc id\nfrom a Drive URL or a google-drive search.\n",
"allowed_tools": [
"Bash"
]
}
Drive Google Docs via curl + jq. The user's OAuth bearer token is in
$GOOGLE_DOCS_TOKEN; every call needs Authorization: Bearer $GOOGLE_DOCS_TOKEN. Base URL: https://docs.googleapis.com/v1/documents. The
token carries documents.readonly (+ identity); writes need documents.
Failures are {"error":{"code","message","status"}} — show verbatim. 401 =
re-install. 403 PERMISSION_DENIED on a write = read-only scope.
The doc id is the …/document/d/<ID>/edit segment of the URL.
D="https://docs.googleapis.com/v1/documents"; AUTH="Authorization: Bearer $GOOGLE_DOCS_TOKEN"
# Read the document. The body is a tree of structural elements; pull plain text:
curl -sS -H "$AUTH" "$D/DOC_ID" \
| jq -r '.title, ([.body.content[]?.paragraph?.elements[]?.textRun?.content] | join(""))'
Create & edit
# Create an empty doc
curl -sS -X POST -H "$AUTH" -H "Content-Type: application/json" \
-d '{"title":"Meeting notes 2026-06-21"}' "$D" | jq '{documentId, title}'
# Insert text at the start (index 1) via batchUpdate (confirm first)
curl -sS -X POST -H "$AUTH" -H "Content-Type: application/json" \
-d '{"requests":[{"insertText":{"location":{"index":1},"text":"Hello\n"}}]}' \
"$D/DOC_ID:batchUpdate" | jq '.documentId'
# Replace all occurrences of a placeholder
curl -sS -X POST -H "$AUTH" -H "Content-Type: application/json" \
-d '{"requests":[{"replaceAllText":{"containsText":{"text":"{{name}}","matchCase":true},"replaceText":"Alex"}}]}' \
"$D/DOC_ID:batchUpdate" | jq '.replies'
Gotchas
- The document model is index-based: text edits target character indices, and every insert shifts later indices — for multiple inserts, apply them back-to-front or recompute indices between calls.
- To get clean Markdown/plain text, walk
body.content[].paragraph.elements[] .textRun.content(the jq above). Tables / lists need deeper traversal. - Creating a doc puts it in the user's Drive root; use the
google-driveskill to move/share it.
Version History
- e0220ca Current 2026-07-05 22:53
