rags
Use this repo skill for RAGs, a Streamlit app that builds configurable LlamaIndex RAG agents from natural-language setup, data sources, and model settings.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- rags
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Use this repo skill for RAGs, a Streamlit app that builds configurable LlamaIndex RAG agents from natural-language setup, data sources, and model settings.
- disable-model-invocation
- true
02 资源文件 6
- 参考文档references/app-architecture.md3.4 KB
- 参考文档references/repo-provenance.md2.3 KB
- 参考文档references/repo-routing-metadata.json487 B
- 参考文档references/troubleshooting.md3.4 KB
- 脚本scripts/check_install.py5.9 KB
- 脚本scripts/run_rags_app.py3.3 KB
03 SKILL.md 正文
RAGs Repo Skill
RAGs is a Streamlit application for building retrieval-augmented chat agents over user-provided data. Use this skill when the task mentions RAGs, creating a RAG bot from natural language, configuring generated RAG parameters, querying a generated agent, or debugging RAGs cache/secrets/model behavior.
Before Acting
- Read
references/repo-provenance.mdwhen checking whether this skill is current for a checkout or when refreshing it. - Read
references/app-architecture.mdwhen you need the cross-page architecture, state, cache, model, and tool flow. - Run or inspect
scripts/check_install.pyfor a safe dependency/source-import diagnostic. It does not call external LLMs. - Use
scripts/run_rags_app.pyto validate or wrap the Streamlit launch command for a user-provided RAGs checkout. It dry-runs by default; pass--executeonly when the user wants a long-running server.
The current source snapshot is an app-style repository, not an installable
Python package named rags. Dependency-oriented setup and running from a RAGs
checkout is the supported operating model captured by this skill.
Route Map
| User intent | Read |
|---|---|
| Build a new RAG bot from files, a directory, URLs, task text, RAG parameters, optional web search, or beta multimodal setup. | sub-skills/builder/SKILL.md |
| Inspect, edit, update, delete, rename, or repair a generated agent's configuration or cache. | sub-skills/configuration/SKILL.md |
| Ask questions to an existing generated agent, inspect sources, or debug no/irrelevant/broken sources. | sub-skills/chat/SKILL.md |
| Diagnose install, secrets, dependency version, root-package install, cache upgrade, or optional dependency problems. | references/troubleshooting.md |
Install and Launch Checks
Use public project instructions or equivalent dependency installation. The
verified inspection environment used Python 3.10 with streamlit==1.28.0,
llama-index==0.9.7, llama-hub==0.0.44, langchain==0.0.305, and
pypdf==3.17.1. A dependency-only Poetry setup or requirements-based setup may
be needed because root package installation fails for this snapshot.
Safe checks:
python scripts/check_install.py
python scripts/check_install.py --repo-root /path/to/rags
python scripts/run_rags_app.py --repo-root /path/to/rags --check-secrets
Launch only with user intent:
python scripts/run_rags_app.py --repo-root /path/to/rags --execute -- --server.headless true
RAGs reads a Streamlit secret named openai_key while configuring the builder
LLM. Provider-specific routes may also need anthropic_key, replicate_key, or
metaphor_key.
Verification Scope
The generated skill is based on source inspection plus live dependency/source
module inspection. Safe checks covered imports, signatures, RAGParams defaults,
local text load_data, Streamlit CLI help, and cache-registry behavior. The
following were intentionally not executed by default: real OpenAI/Anthropic/
Replicate/Metaphor calls, URL downloads, a long-running Streamlit server, and
actual beta multimodal construction with torch/CLIP dependencies.
Boundaries
Do not use this skill as a generic LlamaIndex manual. It is specifically for the RAGs app's builder/configuration/chat workflow and its cache/secrets behavior. For changing the RAGs source code itself, treat this as repository maintenance and combine with ordinary code inspection rather than relying only on this operating 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).