Tiangong KB Patent Search

Use this skill for Tiangong patent-source retrieval. It is intentionally single-source: always search patent, never all, sci, or report.

In Auto Research, a patent-source requirement must name the reviewed patent capability in requiredCapabilityIds; Brave or SCI results cannot stand in for it. If that capability is absent, stop at preflight and have the owner import, configure, license, credential, lock, and smoke the exact patent capability. The standalone wrapper never reads the workspace broker credential store.

Prerequisites

• The standalone wrapper defaults to its separately tested CLI version 0.0.62; this is not the Auto Research workspace runtime. Set TIANGONG_AI_CLI or TIANGONG_AI_CLI_BIN only to intentionally override the standalone entrypoint.
• Set TIANGONG_PATENT_APIKEY or the common fallback TIANGONG_AI_APIKEY. Credentials are rejected recursively in wrapper JSON, request files, URL query parameters, and CLI arguments.
• When request_file / input_file is provided, the wrapper loads .env from that file's directory by default. env_file can point to a different dotenv file. It must be owner-only, regular, and non-symlinked; only documented Tiangong patent credential, endpoint, region, and timeout variables are loaded, and existing environment values win.
• Optionally set TIANGONG_AI_API_BASE_URL; the CLI accepts a Supabase project root, /functions/v1, or /rest/v1 and derives Functions URLs.

Search

For normal searches, pass a query:

./scripts/patent_search.sh '{
  "query": "mechanical recycling polymer purification patent",
  "top_k": 5
}'

When the current directory is inside a managed workspace, this command is blocked before credential loading or network. Only a user who explicitly asks for one isolated query may add "execution_mode":"standalone"; the wrapper then emits a non-secret audit event.

The script calls:

npx --yes --package "@tiangong-ai/cli@0.0.62" -- tiangong-ai research search --query <query> --sources patent --json

For exact edge-function payloads, provide request_file or input_file:

./scripts/patent_search.sh '{
  "request_file": "./patent-request.json",
  "dry_run": true
}'

Raw Payload Filters

Wrapper JSON can include inline raw patent_search fields; the wrapper will forward them through the CLI --input path. The same payload can also be put in request_file / input_file:

{
  "query": "battery recycling equipment",
  "filter": {
    "assignee": ["example company"]
  },
  "datefilter": {
    "publication_date": { "gte": 1262304000 }
  },
  "topK": 5
}

• filter: metadata term filters, shaped as { "field": ["value"] }.
• datefilter: numeric range filters, shaped as { "field": { "gte"?: number, "lte"?: number } }, for indexed numeric/date metadata fields.
• topK: raw edge-function name for result count.
• extK and getMeta are not supported by patent_search.

Input Fields

• query, input, or claim: convenience query text.
• request_file or input_file: JSON body forwarded unchanged.
• env_file: optional dotenv file. Without it, request_file / input_file causes the wrapper to load .env from that file's directory.
• execution_mode: only standalone, and only after an explicit isolated owner request when a managed workspace is detected.
• filter, datefilter, topK: optional inline raw payload fields for patent search.
• sources: optional compatibility field; only patent or default is accepted.
• dry_run: true to return the exact request plan with masked credentials.
• api_base_url; credentials must come from allowed owner environment variables or an owner-only env file.
• patent_url, region, timeout.
• top_k: only used in query mode.