Tiangong Kb Course Search
Search Tiangong knowledge-base course sources through the Tiangong AI CLI.
by tiangong-ai · MIT
The skill
Search Tiangong knowledge-base course sources through the Tiangong AI CLI. Use for courseware, lessons, teaching material, Qinghua/Tsinghua humanities archive course collections, and evidence-grounded education-history retrieval. This skill searches only the course source.
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. SetTIANGONG_AI_CLIorTIANGONG_AI_CLI_BINonly to override the CLI entrypoint intentionally. - Set
TIANGONG_EDUCATION_BEARER_TOKENfor scoped bearer auth, or useTIANGONG_COURSE_APIKEY/TIANGONG_AI_APIKEY. Credentials are never accepted in wrapper JSON, request files, URLs, or CLI arguments. - When
request_file/input_fileis provided, the wrapper loads.envfrom that file's directory by default.env_filecan 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/v1and 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 astagsandraw_relative_pathwhen those keys are indexed for the collection.topK,extK: raw edge-function names for result count and adjacent chunk expansion.datefilteris not exposed by this wrapper for course search.
Input Fields
queryorinput: convenience query text.request_fileorinput_file: JSON body forwarded unchanged.env_file: optional dotenv file. Without it,request_file/input_filecauses the wrapper to load.envfrom that file's directory.filter,topK,extK: optional inline raw payload fields for course search.sources: optional compatibility field; onlycourseordefaultis accepted.dry_run: true to return the exact request plan with masked credentials.api_base_url,course_url,region,timeoutas 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.
Use this skill
Connect your agent to the Skill Store MCP server and ask for it by ref: tiangong-ai/tiangong-kb-course-search-75875634875.
Details
- Publisher
- tiangong-ai
- Category
- Science & Education › Teaching & learning
- License
- MIT
- Updated
- 2026-09-14
References
More from tiangong-ai
- Academic Paper Download Fetch and atomically save structurally and identity-verified academic-paper PDFs from legal open-…
- Ai Tech Fulltext Fetch Fetch and persist article full text for RSS entries already stored in SQLite by ai-tech-rss-fetch.
- Ai Tech Rss Fetch Subscribe to AI and tech RSS feeds and persist normalized metadata into SQLite using mature Pytho…
- Ai Tech Summary Retrieve time-windowed RSS evidence from SQLite and let the agent produce final summaries using R…