AutoRAG-Research
AGENTS.md149 starsChanged 8 months agoApache-2.0View on GitHub
This document defines how the main Codex agent orchestrates specialized sub-agents for multi-step workflows. Each workflow has mandatory human checkpoints where the user must approve before proceeding. When asked to implement a new RAG pipeline from a research paper: When asked to ingest a new dataset into AutoRAG-Research:
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,439 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository9 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- autorag-querySkill
- create-generation-pluginSkill
- create-ingestor-pluginSkill
- create-metric-pluginSkill
- create-retrieval-pluginSkill
- refactorSkill
- resolve-conversationSkill
The file
We split this AGENTS.md into the rules it gives an agent
8 rules. About 0 are common, 4 are distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 7
When asked to implement a new RAG pipeline from a research paper:
- Distinctive: line 20
If the user requests changes, re-invoke the agent with feedback.
- Very specific: line 40
Write a config YAML in configs/ for the new pipeline.
- Very specific: line 57
When asked to ingest a new dataset into AutoRAG-Research:
- Very specific: line 75
Output: Test file in tests/autorag_research/data/
- Distinctive: line 95
Human checkpoints are mandatory: - Never skip approval steps.
- Very specific: line 97
Intermediate files are local only: - Never commit Pipeline_Analysis.json, Pipeline_Design.md, Source_Data_Profile.json, or …
- Distinctive: line 98
Loop on changes: - Keep revising until the user approves.
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
AutoRAG-Research Agent Workflows
This document defines how the main Codex agent orchestrates specialized sub-agents for multi-step workflows. Each workflow has mandatory human checkpoints where the user must approve before proceeding.
Pipeline Implementation Workflow
When asked to implement a new RAG pipeline from a research paper:
Phase 1: Paper Analysis
Spawn pipeline-paper-analyst agent with the paper URL/path.
- Input: Research paper (arxiv link, PDF path, or URL)
- Output:
Pipeline_Analysis.jsonin project root - Present a summary of the extracted algorithm to the user.
Phase 2: Architecture Design (HUMAN CHECKPOINT)
Spawn pipeline-architecture-mapper agent.
- Input:
Pipeline_Analysis.json - Output:
Pipeline_Design.mdin project root - Present the design to the user. Wait for explicit approval before proceeding.
- If the user requests changes, re-invoke the agent with feedback.
Phase 3: Test Writing - TDD (HUMAN CHECKPOINT)
Spawn pipeline-test-writer agent.
- Input: Approved
Pipeline_Design.md - Output:
tests/autorag_research/pipelines/[type]/test_[name]_pipeline.py - Present the tests to the user. Wait for explicit approval before proceeding.
Phase 4: Implementation
Spawn pipeline-implementer agent.
- Input: Approved test file +
Pipeline_Design.md - Output: Pipeline class, config dataclass,
__init__.pyexports
Phase 5: Validation (HUMAN CHECKPOINT)
Spawn pipeline-validator agent.
- Verify against:
ai_instructions/pipeline_checklist.md(all sections must pass) - Output: Validation report (test results +
make checkresults + checklist status) - Present results to the user. Wait for approval before committing.
Phase 6: Wrap Up
- Write a config YAML in
configs/for the new pipeline. - Remind user to delete intermediate files (
Pipeline_Analysis.json,Pipeline_Design.md).
Quick Reference
| Phase | Agent | Output | Human Checkpoint |
| 1 | pipeline-paper-analyst | Pipeline_Analysis.json | No (summary shown) |
| 2 | pipeline-architecture-mapper | Pipeline_Design.md | YES |
| 3 | pipeline-test-writer | test_[name]_pipeline.py | YES |
| 4 | pipeline-implementer | [name].py + config | No |
| 5 | pipeline-validator | Validation report | YES |
Data Ingestor Workflow
When asked to ingest a new dataset into AutoRAG-Research:
Phase 1: Dataset Inspection
Spawn dataset-inspector agent with the dataset name/URL.
- Input: HuggingFace dataset identifier or URL
- Output:
Source_Data_Profile.jsonin project root - Present a summary of the dataset structure to the user.
Phase 2: Schema Design (HUMAN CHECKPOINT)
Spawn schema-architect agent.
- Input:
Source_Data_Profile.json+ai_instructions/db_schema.md - Output:
Mapping_Strategy.mdin project root - Present the mapping strategy to the user. Wait for explicit approval.
- If the user requests changes, re-invoke the agent with feedback.
Phase 3: Test Writing
Spawn test-writer agent.
- Input: Approved
Mapping_Strategy.md - Output: Test file in
tests/autorag_research/data/
Phase 4: Implementation
Spawn implementation-specialist agent.
- Input:
Mapping_Strategy.md+ test file - Output: Ingestor class in
autorag_research/data/
Quick Reference
| Phase | Agent | Output | Human Checkpoint |
| 1 | dataset-inspector | Source_Data_Profile.json | No (summary shown) |
| 2 | schema-architect | Mapping_Strategy.md | YES |
| 3 | test-writer | test file | No |
| 4 | implementation-specialist | ingestor class | No |
Critical Rules
- Human checkpoints are mandatory - Never skip approval steps.
- TDD approach - Tests are always written BEFORE implementation.
- Intermediate files are local only - Never commit
Pipeline_Analysis.json,Pipeline_Design.md,Source_Data_Profile.json, orMapping_Strategy.md. - Loop on changes - Keep revising until the user approves.
- Sequential phases - Each phase depends on the previous one completing successfully.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.