retrieval-client
Guides PageIndexClient workspace persistence, document metadata, structure retrieval, page content lookup, and agentic RAG tools.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- retrieval-client
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Guides PageIndexClient workspace persistence, document metadata, structure retrieval, page content lookup, and agentic RAG tools.
- disable-model-invocation
- true
02 资源文件 1
- 脚本scripts/workspace_smoke.py2.3 KB
03 SKILL.md 正文
Retrieval Client
Use this sub-skill when the user mentions PageIndexClient, workspace directories, document ids, get_document, get_document_structure, get_page_content, cached JSON documents, or agentic vectorless RAG.
Do not use it for initial tree construction unless the user specifically wants PageIndexClient.index():
- classic PDF tree extraction ->
../pdf-indexing/ - Flash PDF tree extraction ->
../flash-indexing/ - Markdown tree extraction ->
../markdown-indexing/
Required context
Read these shared references as needed:
../../references/workflows.md#workspace-retrieval-and-agentic-qafor workflow recipes.../../references/api-reference.md#retrieval-client-and-low-level-toolsfor verified signatures.../../references/data-formats.md#workspace-layoutfor workspace JSON structure.../../references/configuration.md#credentialsfor model key behavior.../../references/troubleshooting.md#workspace-retrievalfor common retrieval failures.
Operating flow
-
Create or load a workspace:
from pageindex import PageIndexClient client = PageIndexClient(workspace="workspace") -
Index only when the user wants to build or refresh a document entry:
doc_id = client.index("document.pdf")index()calls the PDF or Markdown tree builders and can require model credentials. -
Retrieve in the standard order for agent workflows:
print(client.get_document(doc_id)) print(client.get_document_structure(doc_id)) print(client.get_page_content(doc_id, "5-7")) -
For an offline workspace smoke check, run
python scripts/workspace_smoke.pyfrom this sub-skill directory.
Agentic RAG pattern
The repo's agentic demo pattern exposes three tools:
get_document()— confirm document metadata and page or line count.get_document_structure()— inspect the tree without text fields.get_page_content(pages)— fetch tight ranges such as"5-7"or"12".
Keep agent instructions strict:
- Call metadata first.
- Use structure to decide relevant ranges.
- Fetch only tight ranges; never fetch the entire document by default.
- Answer only from tool outputs.
The OpenAI Agents SDK route needs the optional openai-agents dependency plus the user's model credentials. Loading an existing workspace and calling retrieval helpers can be fully offline.
Common decisions
- If a workspace JSON is corrupt, rebuild it by re-indexing or repair
_meta.jsonplus the document JSON file. - If
get_page_contenterrors, validate the page selector syntax before rerunning. - If a Markdown document is queried, explain that ranges refer to heading line numbers rather than PDF pages.
- If the user wants improved extraction quality before retrieval, route back to the relevant indexing sub-skill.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).