Knowhere

Use the Knowhere OpenClaw plugin to ingest local files or URLs, search stored documents, inspect parsed results, check jobs, and clean up stored document state.

Security Scan

The skill is internally consistent: it is an instruction-only wrapper that tells an agent how to use the external Knowhere OpenClaw plugin and requests no unrelated credentials or installs.

Purpose & Capability

Name, description, and runtime instructions consistently describe using the Knowhere OpenClaw plugin (known toolnames like knowhere_ingest_document, knowhere_grep, etc.). The single required config path (plugins.entries.knowhere.enabled) matches the stated prerequisite to enable the plugin.

Instruction Scope

SKILL.md limits runtime actions to calls to knowhere_* tools and guidance on how to pass file paths, URLs, and docIds. It does not instruct the agent to read unrelated system files, environment variables, or to send data to arbitrary endpoints. It does assume the external plugin is installed and enabled.

Install Mechanism

There is no install spec and no code files; the skill is instruction-only. The README and SKILL.md instruct installing the external plugin from a known package name; no archives or third-party download URLs are embedded in the skill itself.

Credentials

The skill declares no environment variables or credentials. The only configuration requirement is enabling the plugin entry (plugins.entries.knowhere.enabled), which is appropriate for a plugin-focused skill.

Persistence & Privilege

The skill does not request always:true and does not modify other skills. Model invocation is enabled (default), which is expected for a tool skill; user-invocable is false (the skill is intended to be used programmatically by agents), which is unusual but not incoherent.

Assessment

This skill itself is just documentation for using the external Knowhere OpenClaw plugin and requests nothing unexpected. Before enabling or using it: (1) install and review the referenced plugin (@ontos-ai/knowhere-claw) from its GitHub repo to confirm you trust its behavior (look for network calls, external uploads, or use of credentials); (2) confirm you are comfortable enabling plugins.entries.knowhere.enabled in your OpenClaw config; (3) be aware agents may call the knowhere tools autonomously (disable model-invocation or restrict the skill if you don't want automatic access); and (4) when attaching files, note the skill expects absolute file paths—ensure the agent has only the file-system access you intend. If you cannot review the plugin code, test in an isolated environment first.

Knowhere Skill

This ClawHub skill depends on the Knowhere OpenClaw plugin and teaches agents how to use the knowhere_* tools well.

Prerequisite

Install the plugin before using this skill:

openclaw plugins install @ontos-ai/knowhere-claw

Then enable the knowhere plugin entry in OpenClaw and restart OpenClaw.

When to Use Knowhere

Use Knowhere when the user wants to:

• ingest a local file or URL into the current scope
• inspect, summarize, or quote previously ingested documents
• inspect ingest jobs or import a completed Knowhere job
• preview, list, remove, or clear stored documents
• understand what fields exist inside the stored result package

Do not assume an uploaded attachment was already ingested. If the user asks you to use an attached file and no existing Knowhere result already covers it, call knowhere_ingest_document.

Tool Selection

• knowhere_ingest_document for new local files or URLs
• knowhere_list_documents to find candidate docId values in the current scope
• knowhere_preview_document for a quick structural overview
• knowhere_grep for text search across chunk fields
• knowhere_read_result_file for manifest.json, hierarchy.json, kb.csv, table HTML, or other text-like files under result/
• knowhere_list_jobs, knowhere_get_job_status, and knowhere_import_completed_job for async jobs
• knowhere_remove_document and knowhere_clear_scope for cleanup

After ingesting a document, use the returned identifiers for follow-up operations instead of guessing names.

Recommended Workflow

1. Ingest or import the document if it is not already in the store.
2. Call knowhere_list_documents if you need to confirm the right docId.
3. Call knowhere_preview_document to get a structural overview.
4. Call knowhere_grep with conditions: [{ pattern: "your query" }] for the default text search path.
5. Narrow by chunk.path, chunk.type, or other conditions when needed.
6. Call knowhere_read_result_file for manifest.json, hierarchy.json, kb.csv, or table HTML when the answer depends on raw package data.

Response Style

Keep tool-driven replies short and labeled.

• Reuse labels such as Scope, Source, File, Chunks, Job ID, and Next
• Prefer one short status line plus the key fields the user needs for the next step
• Keep path in your reasoning and answers when possible
• Cite chunkId and path when answering from retrieved chunks

Retrieval Rules

• Prefer knowhere_grep for text search
• Use knowhere_preview_document before broad reads when the document is large or the relevant branch is unclear
• For image or table questions, inspect matching image or table chunks and related manifest asset entries before answering
• Do not rely on full.md alone when the question depends on exact structure, tables, or images
• When a tool response contains truncatedStrings: true, retry with a higher maxStringChars before answering

Attachment Markers

When a prompt contains a marker like:

[media attached: /absolute/path/to/file.pdf (application/pdf) | handbook.pdf]

Use:

• the exact absolute path as filePath
• the visible filename as fileName

Tool Usage Examples

Ingest a local file:

{
  "filePath": "/tmp/uploads/handbook.pdf",
  "fileName": "handbook.pdf"
}

Ingest a URL:

{
  "url": "https://example.com/report-2026.pdf",
  "title": "Q1 Report"
}

Preview a document:

{
  "docId": "handbook-1234"
}

Search across chunk fields:

{
  "docId": "paper-pdf-a370ef58",
  "conditions": [{ "pattern": "npm audit" }]
}

Read manifest JSON:

{
  "docId": "handbook-1234",
  "filePath": "manifest.json",
  "mode": "json"
}