Imported from tiangong-ai/agent-skills (tiangong-kb-course-search/SKILL.md). Install upstream with npx skills add tiangong-ai/agent-skills --skill tiangong-kb-course-search. Copyright stays with the author.

Tiangong KB Course Search

Use this skill for Tiangong course-source retrieval. It is intentionally single-source: always search course, never all, edu, or textbook.

Prerequisites

• The wrapper defaults to the exact reviewed entrypoint npx --yes --package "@tiangong-ai/cli@0.0.62" -- tiangong-ai; users do not need a preinstalled CLI. Set TIANGONG_AI_CLI or TIANGONG_AI_CLI_BIN only to override the CLI entrypoint intentionally.
• Set TIANGONG_EDUCATION_BEARER_TOKEN for scoped bearer auth, or use TIANGONG_COURSE_APIKEY / TIANGONG_AI_APIKEY. Credentials are never accepted in wrapper JSON, request files, URLs, or 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 an owner-only regular non-symlink file (chmod 600), and only documented Tiangong variables are loaded. Existing process variables 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/course_search.sh '{
  "query": "马约翰 清华 体育 孙立人 梁实秋 钱三强",
  "top_k": 8,
  "ext_k": 20
}'

The script calls:

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

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

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

The script calls:

npx --yes --package "@tiangong-ai/cli@0.0.62" -- tiangong-ai education search --input <request.json> --sources course --json

Raw Payload Filters

Wrapper JSON can include inline raw course_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": "马约翰 清华 体育",
  "filter": {
    "tags": ["thu_humanities"]
  },
  "topK": 8,
  "extK": 20
}

• filter: metadata term filters, shaped as { "field": ["value"] }. Course collections currently use metadata keys such as tags and raw_relative_path when those keys are indexed for the collection.
• topK, extK: raw edge-function names for result count and adjacent chunk expansion.
• datefilter is not exposed by this wrapper for course search.

Input Fields

• query or input: 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.
• filter, topK, extK: optional inline raw payload fields for course search.
• sources: optional compatibility field; only course or default is accepted.
• dry_run: true to return the exact request plan with masked credentials.
• api_base_url, course_url, region, timeout as non-secret routing values.
• top_k, ext_k: only used in query mode.

When writing reader-facing prose from retrieved results, state historical facts directly and cite source titles when useful. Do not mention retrieval mechanics, source grouping, or internal confidence checks.