acprkacprk/crypto-research-agents/AGENTS.md

crypto-research-agents

AGENTS.md15 starsChanged 8 days agoMITView on GitHub

This repo is a toolkit, not a research project. Research happens in a separate project directory initialised with /cra-init <dir> (see docs/QUICKSTART.md). When working in a research project that uses this system: 1. Read STATE.md first. The pi-orchestrator agent (agents/pi-orchestrator.md) owns phase changes. 2. Hand work between agents only through blackboard files (templates/blackboard/). 3. Invariants (never break): - A statement enters paper/ only if its CLAIMS.md status is survived (or weakened, using the weakened wording). - A number enters paper/…

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 31,774 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 repository29 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- abstract-craftSkill
- anonymize-checkSkill
- artifact-packSkill
- baseline-pinSkill
- bench-protocolSkill
- bib-verifySkill
- camera-readySkill
- eprint-searchSkill
- experiments-writingSkill
- falsifySkill
- figure-harvestSkill
- idea-mining-loopSkill
- lean-bridgeSkill
- lit-matrixSkill
- log-to-evidenceSkill
- mock-reviewSkill
- paper-figuresSkill
- paper-playbookSkill
- param-estimationSkill
- phase-gateSkill
- polish-writingSkill
- reading-notesSkill
- rebuttalSkill
- related-work-writingSkill
- sage-checkSkill
- technique-overviewSkill
- textbook-indexSkill
- venue-calibrationSkill

The file

We split this AGENTS.md into the rules it gives an agent

4 rules. About 0 are common, 1 is distinctive, and 3 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.
3
Very specific
Tied to their project's names, files or tools. Probably skip.

Common
Distinctive
Very specific

Distinctive:
When working *in a research project* that uses this system:
line 6
Very specific:
Read STATE.md first. The pi-orchestrator agent (agents/pi-orchestrator.md) owns phase changes.
line 7
Very specific:
Prefer lib/cryptomath + lib/bench over ad-hoc scripts.
line 14
Very specific:
When working *on this repo*: follow SPEC.md (names, formats, the no-private-content rule).
line 17

≈ “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.

Keep what fits and drop the rest. Your version, ready for your agents.

Files

AGENTS.md

crypto-research-agents — instructions for Codex / generic coding agents

This repo is a toolkit, not a research project. Research happens in a separate project directory initialised with /cra-init <dir> (see docs/QUICKSTART.md).

When working in a research project that uses this system:

Read STATE.md first. The pi-orchestrator agent (agents/pi-orchestrator.md) owns phase changes.
Hand work between agents only through blackboard files (templates/blackboard/).
Invariants (never break):
- A statement enters paper/ only if its CLAIMS.md status is survived (or weakened, using the weakened wording).
- A number enters paper/ only if it has an EVIDENCE.md row; run crbench audit-tex paper/main.tex EVIDENCE.md.
- A bib entry stays only if skills/bib-verify confirmed it exists.
- Nothing leaves the machine (submission, e-mail, eprint, public repo) without explicit human approval.
Prefer lib/cryptomath + lib/bench over ad-hoc scripts; put throw-away scripts under falsify/<claim-id>/ or results/<exp>/.
Toy implementations in lib/cryptomath/fhe|lattice|protocols are insecure and only for reasoning / small experiments.

When working on this repo: follow SPEC.md (names, formats, the no-private-content rule).

Using the system through MCP (recommended outside Claude Code)

cra-mcp install --host <codex|cline|gemini|cursor|...> --project <dir> connects the cra MCP server (see docs/integrations/README.md). Then call the cra_resume tool first in every session and change the ledgers only through the typed tools (claim_register, claim_verdict, evidence_add, decision_log, gate_tick, phase_advance), which enforce the invariants above.

Using the agents without Claude Code subagents

Each file in agents/ is a self-contained role prompt. In tools without subagents, "become" an agent by reading its file and following its Procedure; skills in skills/<name>/SKILL.md are plain Markdown procedures with optional scripts. Slash commands in commands/ are equivalent to typing their body as a prompt.

Download now (ZIP)

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

What did you use it for? e.g. Reviewing pull requests in a Rails app

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.

Check your setup, free