memos（memos 搜索）

Memos API Skill

This skill combines two layers:

• low-level API access for exact Memos endpoints
• task-oriented query helpers for common memo retrieval jobs
• tag inventory helpers so the calling agent can inspect existing hashtags before deciding whether to reuse them

Scope is intentionally limited to:

• attachment operations
• memo operations
• activity operations

It does not cover user management, auth setup beyond the local .env, shortcuts, identity providers, or instance settings. If the request drifts there, say the skill scope is limited instead of guessing unsupported commands.

What this skill provides

• A Bun CLI at scripts/memos.ts
• Endpoint discovery and direct API calling via ops, describe, and call
• Task-oriented memo helpers via latest, recent, and search
• A list-tags helper that aggregates existing tags, counts, recency, and sample memo snippets
• A file-to-attachment helper so agents do not have to hand-roll base64 payloads
• A compact operation catalog at references/api-summary.md
• Query recipes and time-window guidance at references/query-recipes.md

Setup

1. Work from the skill directory.
2. Ensure .env contains:

- MEMOS_BASE_URL - MEMOS_ACCESS_TOKEN

1. Verify config:

bun run scripts/memos.ts config

MEMOS_BASE_URL may be either the instance root such as http://localhost:5230 or the API base such as http://localhost:5230/api/v1. The CLI normalizes both.

Choose the right path

Use the high-level commands first when the user asked for a retrieval task, not a specific endpoint.

• Latest memo or latest N memos:

bun run scripts/memos.ts latest --count 5

• Recent memos inside a time window:

bun run scripts/memos.ts recent --days 7 --window rolling --include-content

• Search by text or tag:

bun run scripts/memos.ts search --text agent --tag Thought --days 30

• Inspect existing tags before deciding which hashtags to reuse in a new memo:

bun run scripts/memos.ts list-tags --limit 100 --sample-size 2

Use call when the user explicitly needs one documented endpoint or a write operation.

• Inspect the exact endpoint first:

bun run scripts/memos.ts describe MemoService_UpdateMemo

• Then call it directly:

bun run scripts/memos.ts call MemoService_UpdateMemo \
  --path memo=YOUR_MEMO_ID \
  --query updateMask=content,visibility \
  --body '{"content":"Updated content","visibility":"PROTECTED"}'

Known quirks

• call --paginate returns {operationId, pageCount, pages: [...]}, not a top-level {memos: [...]}. Flatten pages[*].memos before post-processing.
• Memo time checks should use displayTime first, then fall back to createTime.
• For time-range queries, prefer orderBy='display_time desc' plus local filtering. Do not assume the server-side filter field supports stable time semantics across versions.
• The task-oriented commands default state=NORMAL so archived or deleted content does not silently leak into user-facing summaries.
• search --tag matches the memo tags array exactly. It is not a full-text hashtag parser over content.
• Memo.tags is extracted by the server from hashtags in content. Treat it as output-only unless the deployed API docs explicitly say otherwise.

Recommended workflow

1. Retrieval tasks

When the user asks questions like:

• 最近一周有哪些备忘录
• 最新的一条 memo 是什么
• 带 Thought tag 的 memo 有哪些
• 搜一下包含某个关键词的 memo

Use:

1. latest, recent, or search
2. Return the important fields from the JSON result
3. Summarize content for the user when they asked for meaning, not raw JSON

Read references/query-recipes.md when you need concrete examples or want to choose between rolling and calendar time windows.

2. Tag-aware memo creation

When the user asks to create a memo with tags or hashtags:

1. Run bun run scripts/memos.ts list-tags... first to inspect the existing tag vocabulary.
2. If one or more candidate tags might match, run search --tag... on the likely choices to inspect prior memo context.
3. Let the calling agent decide which existing tags to reuse. Do not hard-code similarity logic in scripts for this decision.
4. Put the final tags directly into the memo content as hashtags, usually at the top, for example:

{
  "state": "NORMAL",
  "visibility": "PRIVATE",
  "content": "#memos #openclaw\n\nImplemented tag discovery before creating this memo."
}

1. Call MemoService_CreateMemo with that content body.

Prefer reusing an existing tag when it clearly expresses the same concept. If the meaning is uncertain, inspect old memo examples first instead of inventing normalization rules.

3. Long content fallback

When creating or updating a memo, if the content is too long for one memo:

1. Keep the first chunk in the memo itself.
2. Append the overflow to comments with MemoService_CreateMemoComment.
3. If one comment is still too long, keep splitting the remaining overflow into multiple ordered comments until everything is stored.

Use this as the default fallback instead of truncating content silently.

4. Exact API work

When the user asks to create, update, delete, attach files, add reactions, or inspect one exact endpoint:

1. Run bun run scripts/memos.ts ops
2. If needed, run bun run scripts/memos.ts describe <operationId>
3. Use bun run scripts/memos.ts call <operationId>...

This keeps the agent aligned with the documented API surface instead of inventing raw requests.

5. Resource naming rules

Path placeholders usually expect bare ids:

• --path memo=0195c...
• --path attachment=abc123
• --path activity=evt_123

Request bodies often expect full resource names:

• memos/0195c...
• attachments/abc123

If you accidentally pass a full resource name into a path flag, the CLI strips the leading path and keeps the last segment.

6. Update requests

Patch endpoints such as:

• AttachmentService_UpdateAttachment
• MemoService_UpdateMemo

need --query updateMask=....

Without it, many updates fail or update less than expected.

7. Complex bodies and binary uploads

Prefer @file.json for larger bodies:

bun run scripts/memos.ts call MemoService_CreateMemo --body @/tmp/memo.json

Use the helper for attachments instead of hand-building base64:

bun run scripts/memos.ts attachment-body --file /path/to/file --output /tmp/body.json

Then create the attachment:

bun run scripts/memos.ts call AttachmentService_CreateAttachment --body @/tmp/body.json

Read next when needed

• Read references/query-recipes.md for recent/latest/search workflows, time-window semantics, and memo post-processing examples.
• Read references/api-summary.md for the compact endpoint catalog, body hints, and pagination details.

Output expectations

When using this skill for execution:

• call the real API instead of only describing it
• prefer latest, recent, or search for retrieval tasks instead of hand-writing pagination logic each time
• return the important response fields to the user
• if you create or update resources, include the resource name or id in your answer
• if the API returns an error, surface the status and body plainly instead of hiding it