agentleFS
Sign inStart free
herakles-devherakles-dev/typesafe-claude-kit/CLAUDE.md

typesafe-claude-kit

CLAUDE.md2 starsChanged 53 days agoMITView on GitHub

A Claude Code kit for building decision layers on TypeSafe's Jev model: typed Choice / Score / Noul judgments with calibrated probabilities, consumed directly by code. Core rule: code owns the workflow, the model only judges. Jev never generates text, writes code, or picks a next action — it answers one bounded, typed question at a time. 1. AGENTS.md — the portal: 60-second orientation, the fit test, standing rules, routing table. Read this before anything else; it sends you to…

  • Reads credentials

The file

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

5 rules. About 0 are common, 4 are distinctive, and 1 is 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.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read this before anything else.

    line 12
  2. Very specific:

    When unsure, start with typesafe-architect — telling you TypeSafe is the wrong tool is a valid, common answer.

    line 55
  3. Distinctive:

    Every question is one judgment, phrased literally.

    line 60
  4. Distinctive:

    All questions for one piece of state batched into a single request

    line 61
  5. Distinctive:

    No math, counting, dates, or generation sent to the model

    line 62

≈ “Common” is an estimate: at least 1 in 5 of the 26,980 CLAUDE.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
CLAUDE.md

typesafe-claude-kit

A Claude Code kit for building decision layers on TypeSafe's Jev model: typed Choice / Score / Noul judgments with calibrated probabilities, consumed directly by code.

Core rule: code owns the workflow, the model only judges. Jev never generates text, writes code, or picks a next action — it answers one bounded, typed question at a time.

Where to start reading

  1. AGENTS.md — the portal: 60-second orientation, the fit test, standing rules, routing table. Read this before anything else; it sends you to sections, not whole files.
  2. knowledge/MASTERY.md — the distilled reference, read by section per the routing table.
  3. reference/ — the full upstream doc mirror, fetched (not vendored) by scripts/fetch-docs.sh. Run that once before relying on it; it's gitignored.

Running things

./setup.sh                        # first-time setup: deps, agents/skill, .env, docs mirror
python3 tests/run_all.py --list   # see what suites exist, no API calls
python3 tests/run_all.py          # run everything live (needs TYPESAFE_API_KEY)
python3 tools/question_critic.py --file draft.json          # critique a drafted question
python3 tools/jaggedness_screen.py --file design.json        # screen a design for failure modes
python3 tools/consistency_probe.py --set ticket_triage --inputs examples/labels/ticket_triage.json

Environment

VariableRequiredDefault
TYPESAFE_API_KEYYesnone — env var, or .env at the kit root
TYPESAFE_USAGE_LOGNo~/.typesafe/usage.jsonl
TYPESAFE_KITNo. — set if agents run from outside the clone
TYPESAFE_AGENT_ROSTERNoexamples/agent-roster.json

Never send Jev

Arithmetic · counting (characters, occurrences, list items) · date/time comparison or ordering · text generation. These are documented failure modes (reference/model-jaggedness/jev-1.13.md), not edge cases. Also don't assume P(x) == 1 - P(not x), and never port a threshold between a Noul and a Choice — they don't share a scale.

Which agent to reach for

NeedAgent
Find where TypeSafe fits, or check for upstream drifttypesafe-scout
Decide the shape: code vs. judgment, request topologytypesafe-architect
Write or fix the actual questionstypesafe-question-smith
Implement: QuestionSet, client wiring, live teststypesafe-engineer
Turn an invented threshold into a measured onetypesafe-calibrator
Red-team a design before it shipstypesafe-adversary

When unsure, start with typesafe-architect — telling you TypeSafe is the wrong tool is a valid, common answer. Details and the tool each specialist built: agents/README.md.

Checklist before shipping an integration

  • Every question is one judgment, phrased literally; levels describe situations, not degrees
  • All questions for one piece of state batched into a single request
  • No math, counting, dates, or generation sent to the model
  • Thresholds are named in code and validated on real data before gating anything irreversible
  • Model version pinned (jev-1.13.0) wherever a threshold exists
  • Adversarial inputs tested if the state contains untrusted text

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