wayfinder
AGENTS.md0 starsChanged 4 months agoMITView on GitHub
Companion to CLAUDE.md. Read both. This project is a hackathon Researcher-track entry framed as a researcher who also builds: Wayfinder, a hypothesis referee that validates T-cell mechanistic claims against real Perturb-seq data, returning a verdict with a receipt per hop. The deliverable is a reproducible finding + how Claude Science reached it; the referee / LBD engine / Claude-Science automation are instruments the researcher built to ask and answer the research questions (the tool is the method/vehicle, not a product…
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 33,315 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 repository5 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- atomic-plannerSkill
- freshenSkill
- session-closerSkill
- session-startSkill
The file
We split this AGENTS.md into the rules it gives an agent
3 rules. About 0 are common, 1 is distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- could be automatic: a setting could enforce it
- Distinctive: line 12
Never let an LLM assert biology: Every causal edge must trace to a data receipt (odds ratio / p-value / effect size) from the Perturb-seq …
- Very specific: line 14
Never commit raw data or secrets (see .gitignore).
- Very specific: line 18
Keep changes small and testable.
≈ “Common” is an estimate: at least 1 in 5 of the 42,283 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
AGENTS.md — cross-agent brief (Codex & other agents)
Companion to CLAUDE.md. Read both. This project is a hackathon Researcher-track entry framed as a researcher who also builds: Wayfinder, a hypothesis referee that validates T-cell mechanistic claims against real Perturb-seq data, returning a verdict with a receipt per hop. The deliverable is a reproducible finding + how Claude Science reached it; the referee / LBD engine / Claude-Science automation are instruments the researcher built to ask and answer the research questions (the tool is the method/vehicle, not a product to adopt).
Stack
- Python >=3.11. Package/dep tooling:
uv(orpip+ venv). Data:pandasover local CSVs. - LLM reasoning: Anthropic Claude (Claude SDK). Embeddings (if needed): Voyage AI. Never OpenAI.
- Retrieval/graph (as needed): LlamaIndex / on-disk. UI: Streamlit. Deploy: a simple managed host (written fresh).
Never do
- Never reuse pre-existing project code (new-work-only hackathon). Author everything fresh. The old PyZoBot POC is a mental reference only — do not copy files.
- Never let an LLM assert biology. Every causal edge must trace to a data receipt (odds ratio / p-value / effect size) from the Perturb-seq tables.
- Never use uncalibrated language ("discovered/proven"). Use "consistent with / re-derived / refuted / untested / flagged."
- Never commit raw data or secrets (see
.gitignore). Never add OpenAI as a provider.
Working agreement
- Data-validation lookups are deterministic functions/tools, not LLM calls. Reserve LLM/agent calls for genuine judgment.
- Keep changes small and testable; the builder works in short async bursts. The live handoff is
memory/NEXT_SESSION.md(dashboard:WORK_PROGRESS.md, index:MEMORY.md) — leave it updated. - The 3-hop validation substrate + the DTO/tool contract are in
docs/plan.md(sections 3 and 5). Follow it.
How to run
Fetch data: bash data/fetch_data.sh. (Build/run commands land as the app is written.)
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.

