jev-awesome-skills

# @aitofy/jev-awesome-skills

Full text for language models. Generated by scripts/llms.mjs. Do not edit by hand.


----- README.md -----

# Jev skills for Claude Code, Codex, Cursor, and Grok

[![npm](https://img.shields.io/npm/v/@aitofy/jev-awesome-skills)](https://www.npmjs.com/package/@aitofy/jev-awesome-skills)
[![CI](https://github.com/aitofy-dev/jev-awesome-skills/actions/workflows/ci.yml/badge.svg)](https://github.com/aitofy-dev/jev-awesome-skills/actions/workflows/ci.yml)
[![MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE)

Open-source TypeSafe Jev skills. One client decides when to proceed, when to ask you, and when to stop — then uses that same judgment to route a skill, guard a command, verify a claim, triage a record, review a diff, or pick the next legal action.

Jev (TypeSafe System One) returns probabilities for a closed `choice`, `noul`, or `score`. It does not write the code. This is the catalog a Claude Code, Codex, Cursor, or Grok agent runs, instead of a hand-rolled Jev script per job. Package name: `@aitofy/jev-awesome-skills`.

## Install

```bash
npm i @aitofy/jev-awesome-skills
```

A Git install builds `dist` before the package is packed, so the import and the `jev` command exist without a committed build:

```bash
npm i github:aitofy-dev/jev-awesome-skills
```

Node 20+. ESM. No runtime dependencies. The API key stays in the environment variable `TYPESAFE_API_KEY`. The CLI never prints it. Dry-run and `--fixture` need no key.

Copy the skill folders into the directory your agent already reads. Claude Code uses `~/.claude/skills/`.

```bash
cp -R skills/jev-awesome-skills skills/jev-gate skills/jev-route skills/jev-guard \
  skills/jev-verify skills/jev-triage skills/jev-review skills/jev-act \
  ~/.claude/skills/
```

Prompt you can give the agent:

```text
Install @aitofy/jev-awesome-skills, or the GitHub repo of the same name.
Run the jev command from that install (`npx jev --dry-run --file request.json`). Do not write a second HTTP client.
Dry-run first. Do not call the network until I confirm.
Read the API key only from the environment. Never print it or ask me to paste it.
```

## 30 seconds, no key

```js
import { decideFixture } from '@aitofy/jev-awesome-skills'

const decision = decideFixture({
  model: 'fixture-offline',
  answers: {
    pick: { type: 'choice', choice: 'pnpm', probabilities: { pnpm: 0.97, npm: 0.03 } },
    determined: { type: 'noul', noul: 0.95 },
    harm: {
      type: 'score',
      score: 0.2,
      legend: { 0: 'reversible', 2: 'destructive' },
      probabilities: { 0: 0.8, 2: 0 },
    },
  },
})

console.log(decision.branch) // proceed
```

From a clone, after `npm run build`:

```bash
node examples/decide.mjs
jev --fixture examples/guard.response.json
```

`guard` is a stop. `route` is an ask. `gate` is a proceed. Same cutoffs, three jobs.

## How this differs

| | Chat model decides | Ad-hoc Jev script | Official TypeSafe skill | This package |
| --- | --- | --- | --- | --- |
| What you get | Prose | One curl per task | How to write a call | Seven jobs one CLI can run |
| Proceed / ask / stop | Vibes | You invent the numbers | You invent the numbers | 0.9 / 0.9 / 1.5, stop wins |
| Missing key or a bad body | n/a | Crash, or a made-up number | n/a | `source: "fallback"`, no fake probability |
| Secrets in state | Sent along | Easy to send | Your problem | Redacted before the body exists |
| Who writes code | The agent | You | The agent | The agent. Jev only returns probabilities |

## Catalog

| Skill | Job | Template |
| --- | --- | --- |
| [jev-awesome-skills](skills/jev-awesome-skills/SKILL.md) | When to call, fail open, cutoffs | — |
| [jev-gate](skills/jev-gate/SKILL.md) | The repo may already decide the choice | [gate.request.json](examples/gate.request.json) |
| [jev-route](skills/jev-route/SKILL.md) | Which installed skill to load, or none | [route.request.json](examples/route.request.json) |
| [jev-guard](skills/jev-guard/SKILL.md) | Before a command that might destroy data or leak a secret | [guard.request.json](examples/guard.request.json) |
| [jev-verify](skills/jev-verify/SKILL.md) | Before you report a claim or "done" | [verify.request.json](examples/verify.request.json) |
| [jev-triage](skills/jev-triage/SKILL.md) | One label for a record you already have | [triage.request.json](examples/triage.request.json) |
| [jev-review](skills/jev-review/SKILL.md) | Ship, fix, or ask a person. Not a merge button | [review.request.json](examples/review.request.json) |
| [jev-act](skills/jev-act/SKILL.md) | Next action from a list you observed. No invented clicks | [act.request.json](examples/act.request.json) |

`route` is how this integrates with skills you already installed. Pass their names as `--option`. This repo does not vendor them.

```bash
jev --dry-run --file examples/gate.request.json
jev questions triage
jev --file request.json
```

Live calls go to `POST https://api.typesafe.ai/v1/systemone` with model `jev-latest`. Questions in one request share one state. The host still runs tools and checks results.

## Cutoffs

Stop when `harm.score` is at or above **1.5**. Proceed when the chosen label's probability is at or above **0.9** and `determined.noul` is at or above **0.9**. Otherwise ask. Stop wins. The numbers live in one object and in [the policy skill](skills/jev-awesome-skills/SKILL.md).

Harm levels are: read-only or reversible, testable change, may destroy data or leak a secret or leave the task.

## In the wild

Other public Jev work, listed so you can see the jobs. A link is not an endorsement, a benchmark, or a star count. We have not run these repos.

| Project | What it is |
| --- | --- |
| [typesafe-ai/skills](https://github.com/typesafe-ai/skills) | Official skill for writing Jev calls |
| [wuyoscar/jev-skill](https://github.com/wuyoscar/jev-skill) | Large scenario collection for agents |
| [browser-use/jev-ultrafast](https://github.com/browser-use/jev-ultrafast) | Browser agent, Jev picks the action |
| [tamaratran/fast-jev-compaction](https://github.com/tamaratran/fast-jev-compaction) | Keep or delete history without rewriting it |
| [devagrawal09/jev-review](https://github.com/devagrawal09/jev-review) | Staged code review |
| [DevMortimer/pi-warden](https://github.com/DevMortimer/pi-warden) | Drift and unsupported "done" |
| [Dicklesworthstone/skillranker](https://github.com/Dicklesworthstone/skillranker) | Rank installed skills |
| [ShivamPansuriya/jev-skill-gate](https://github.com/ShivamPansuriya/jev-skill-gate) | Hide skills that do not fit the turn |
| [jkudish/jev-mcp](https://github.com/jkudish/jev-mcp) | Jev as MCP tools |
| [dbreunig/building-with-jev-skill](https://github.com/dbreunig/building-with-jev-skill) | Skill for designing questions |
| [Anil-matcha/awesome-jev-by-typesafe](https://github.com/Anil-matcha/awesome-jev-by-typesafe) | Project directory |

## For agents and crawlers

- [llms.txt](llms.txt) is the index, with raw links.
- [llms-full.txt](llms-full.txt) is the README and every skill in one file.
- Raw index: <https://raw.githubusercontent.com/aitofy-dev/jev-awesome-skills/main/llms.txt>

Regenerate both with `node scripts/llms.mjs` after a skill edit.

## Development

```bash
npm install
npm test
```

MIT. No telemetry.


----- llms.txt -----

# Jev awesome skills

> Open-source TypeSafe Jev (System One) skills for coding agents. Closed choice, noul, and score judgments. The host writes the code. Jev returns probabilities only: proceed, ask, or stop.

People find this as a Jev skill for Claude Code, Codex, Cursor, or Grok, and as an awesome Jev catalog. It is not the official TypeSafe skill, and it does not invent star counts or accuracy numbers.

Install today from Git: `npm i github:aitofy-dev/jev-awesome-skills`. The npm name is `@aitofy/jev-awesome-skills`. Dry-run needs no API key. A live call reads the environment variable TYPESAFE_API_KEY and must not print it.

## Docs

- [README](https://github.com/aitofy-dev/jev-awesome-skills/blob/main/README.md): what it is, install, one offline example. Raw: https://raw.githubusercontent.com/aitofy-dev/jev-awesome-skills/main/README.md
- [llms-full.txt](https://raw.githubusercontent.com/aitofy-dev/jev-awesome-skills/main/llms-full.txt): this index, the README, and every skill in one file.
- [Changelog](https://github.com/aitofy-dev/jev-awesome-skills/blob/main/CHANGELOG.md)

## Skills

- [jev-awesome-skills](https://github.com/aitofy-dev/jev-awesome-skills/blob/main/skills/jev-awesome-skills/SKILL.md): When to call Jev, fail open, and the proceed / ask / stop cutoffs. Raw: https://raw.githubusercontent.com/aitofy-dev/jev-awesome-skills/main/skills/jev-awesome-skills/SKILL.md
- [jev-gate](https://github.com/aitofy-dev/jev-awesome-skills/blob/main/skills/jev-gate/SKILL.md): Proceed, ask, or stop when the repo may already decide a closed choice. Raw: https://raw.githubusercontent.com/aitofy-dev/jev-awesome-skills/main/skills/jev-gate/SKILL.md
- [jev-route](https://github.com/aitofy-dev/jev-awesome-skills/blob/main/skills/jev-route/SKILL.md): Pick one installed skill or tool, or none. Raw: https://raw.githubusercontent.com/aitofy-dev/jev-awesome-skills/main/skills/jev-route/SKILL.md
- [jev-guard](https://github.com/aitofy-dev/jev-awesome-skills/blob/main/skills/jev-guard/SKILL.md): Before a command that might destroy data, leak a secret, or leave the task. Raw: https://raw.githubusercontent.com/aitofy-dev/jev-awesome-skills/main/skills/jev-guard/SKILL.md
- [jev-verify](https://github.com/aitofy-dev/jev-awesome-skills/blob/main/skills/jev-verify/SKILL.md): Before reporting a claim or done. Raw: https://raw.githubusercontent.com/aitofy-dev/jev-awesome-skills/main/skills/jev-verify/SKILL.md
- [jev-triage](https://github.com/aitofy-dev/jev-awesome-skills/blob/main/skills/jev-triage/SKILL.md): One label for a record you already have. Raw: https://raw.githubusercontent.com/aitofy-dev/jev-awesome-skills/main/skills/jev-triage/SKILL.md
- [jev-review](https://github.com/aitofy-dev/jev-awesome-skills/blob/main/skills/jev-review/SKILL.md): Ship, fix, or ask a person. Does not merge. Raw: https://raw.githubusercontent.com/aitofy-dev/jev-awesome-skills/main/skills/jev-review/SKILL.md
- [jev-act](https://github.com/aitofy-dev/jev-awesome-skills/blob/main/skills/jev-act/SKILL.md): Next action from a list you observed. No invented clicks. Raw: https://raw.githubusercontent.com/aitofy-dev/jev-awesome-skills/main/skills/jev-act/SKILL.md

## Examples

- [Gate request](https://raw.githubusercontent.com/aitofy-dev/jev-awesome-skills/main/examples/gate.request.json)
- [Gate fixture, branch proceed](https://raw.githubusercontent.com/aitofy-dev/jev-awesome-skills/main/examples/gate.response.json)
- [Route fixture, branch ask](https://raw.githubusercontent.com/aitofy-dev/jev-awesome-skills/main/examples/route.response.json)
- [Guard fixture, branch stop](https://raw.githubusercontent.com/aitofy-dev/jev-awesome-skills/main/examples/guard.response.json)


----- CHANGELOG.md -----

# Changelog

## 0.1.1

`prepare` builds the package on install from Git. A clone has no `dist/`; the install produces `dist/index.js` and the `jev` command.

## 0.1.0

First public release of the Jev skill catalog for coding agents.

- Build and parse `choice`, `noul`, and `score` requests for TypeSafe System One.
- Redact secret-shaped state before a request body exists.
- Fail open, with no invented probability, when the API key is missing, the transport throws, or the body is not Jev answers.
- One proceed / ask / stop policy, shared by gate, route, guard, verify, triage, review, and act.
- Offline `--dry-run` and `--fixture`. The API key is read only from the environment.


----- skills/jev-awesome-skills/SKILL.md -----

---
name: jev-awesome-skills
description: >
  Use when a coding agent must make a closed choice, noul (yes/no), or score
  judgment with TypeSafe Jev instead of guessing. Load it to proceed, ask, or
  stop; to route a skill; to guard a command; to verify a claim; to triage a
  record; to review a diff; or to pick the next legal action. Jev returns
  probabilities only. The host still writes the code and checks the result.
  Skip it for open-ended writing, or when one file or one command already
  answers the question.
---

# Jev awesome skills

Jev answers a closed question. It does not edit the repo, invent a plan, or replace a test. You write the code and check the result.

Every skill in this package runs `@aitofy/jev-awesome-skills`. Do not write a second client.

## Catalog

| Job | Skill | Use it when |
| --- | --- | --- |
| Policy | jev-awesome-skills | You need the cutoff rules below |
| Gate | jev-gate | A closed choice may already be determined by the repo |
| Route | jev-route | More than one installed skill could handle the turn |
| Guard | jev-guard | A command might destroy data, leak a secret, or leave the task |
| Verify | jev-verify | You are about to report a claim or "done" |
| Triage | jev-triage | A record needs one label |
| Review | jev-review | A diff needs ship, fix, or a person |
| Act | jev-act | The next step must be one of the actions you already listed |

## Ask

```
jev questions gate --option pnpm="pnpm-lock.yaml is present" --option ask_user="not determined"
jev --dry-run --file request.json
jev --file request.json
jev --fixture examples/gate.response.json
```

Dry-run and `--fixture` do not open a socket. A live call reads the API key from the environment variable `TYPESAFE_API_KEY`. Never print the key. Never commit it. Never ask the user to paste it into chat.

## Fail open

If the key is missing, the transport throws, or the body is not Jev answers, the command returns `"source": "fallback"` and no answer probabilities. Fail open: do not invent a probability and call it Jev. Decide yourself, or ask the user.

## Proceed, ask, or stop

Read `branch` when the answers include `pick`, `determined`, and `harm`.

- **stop** when `harm.score` is at or above **1.5**. That is the midpoint between "changes behavior but is testable" and "may destroy data, leak a secret, or go beyond the task". Stop wins. Ask the user before any command that can do that, even if the other probabilities are high.
- **proceed** when harm is below 1.5, the probability of the chosen `pick` label is at or above **0.9**, and `determined.noul` is at or above **0.9**. The context already determines the answer. Do not ask the user. Tell them what you inferred in one line.
- **ask** otherwise. The question is undetermined.

Use the probability of the chosen label, not `confidence`.

If one file or one command already answers it, proceed without calling Jev and without asking. If you already know the command can destroy data, leak a secret, or leave the task, stop for confirmation even before calling Jev.

A proceed branch is not a passing test. You still write the change and run the checks.


----- skills/jev-gate/SKILL.md -----

---
name: jev-gate
description: >
  Use when a closed choice may already be determined by the repo, such as
  package manager, test runner, or file location, and you must proceed, ask
  the user, or stop. Runs @aitofy/jev-awesome-skills. Not for open-ended design.
---

# Gate

You own the options. Jev only says which listed option the context supports.

1. Read the files that could decide the choice. If one of them already settles it, proceed and do not call Jev.
2. Otherwise build the question with every real option plus `ask_user`:

```
jev questions gate --option pnpm="pnpm-lock.yaml is present" --option ask_user="the lockfile does not decide"
```

3. Put the evidence in `state`. Run `jev --dry-run --file request.json`, then `jev --file request.json` from `@aitofy/jev-awesome-skills`. Do not write a second client.
4. Apply `branch`. Stop when `harm.score` is at or above 1.5. Proceed when the chosen label's probability is at or above 0.9 and `determined.noul` is at or above 0.9. Otherwise ask. Stop wins.
5. On `"source": "fallback"`, fail open. Do not invent a probability.

Template: `examples/gate.request.json`. Offline sample: `examples/gate.response.json` returns proceed for pnpm.


----- skills/jev-route/SKILL.md -----

---
name: jev-route
description: >
  Use when more than one installed skill or tool could handle the turn and
  you need Jev to pick one, or none, before loading it. Runs
  @aitofy/jev-awesome-skills. Do not load the whole skill roster into context first.
---

# Route

You list the skills that are actually installed. Jev picks one label. You load that skill and do the work.

Keep `none` and `ask_user` in the criteria. Add each real skill with its one-line description:

```
jev questions route --option repo-review="Inspect the diff and report file-backed findings"
```

Presets already include `none` and `ask_user`. Put the user request and the installed names in `state`. Run `jev` from `@aitofy/jev-awesome-skills`. Do not write a second client.

Stop when `harm.score` is at or above 1.5. Proceed when the chosen label's probability is at or above 0.9 and `determined.noul` is at or above 0.9. Otherwise ask. Stop wins. `none` means answer directly. Do not load a skill the user did not install.

On `"source": "fallback"`, fail open and pick with your own judgment. Do not invent a probability.

Template: `examples/route.request.json`.


----- skills/jev-guard/SKILL.md -----

---
name: jev-guard
description: >
  Use before running a shell command that might delete data, print a secret,
  force-push, or do something the user did not ask for. Jev scores the harm.
  You still decide whether to run it. Runs @aitofy/jev-awesome-skills.
---

# Guard

If you already know the command can destroy data, leak a secret, or leave the task, stop and ask the user. Do not use Jev to talk yourself into it.

Otherwise put the exact command, cwd, and task in `state` and run:

```
jev questions guard
jev --file request.json
```

The package is `@aitofy/jev-awesome-skills`. Do not write a second client. Preset labels are `allow`, `ask_user`, and `refuse`.

Stop when `harm.score` is at or above 1.5. Proceed when the chosen label's probability is at or above 0.9 and `determined.noul` is at or above 0.9. Otherwise ask. Stop wins. Proceed means you may run that exact command, then check the result. It does not permit a second command.

On `"source": "fallback"`, fail open: ask the user before anything hard to undo. Do not invent a probability.

Template: `examples/guard.request.json`. Offline sample: `examples/guard.response.json` returns stop.


----- skills/jev-verify/SKILL.md -----

---
name: jev-verify
description: >
  Use before reporting a claim, a fix, or "done". Jev judges whether the
  evidence you supply supports the claim. It does not run the tests for you.
  Runs @aitofy/jev-awesome-skills.
---

# Verify

Collect the receipts first: failing test before, passing test after, command, exit code. If you have no receipt, do not claim the work is done and do not call Jev to bless the sentence.

Put the claim and the receipts in `state`. Run `jev questions verify`, then `jev --file request.json` from `@aitofy/jev-awesome-skills`. Do not write a second client.

Labels are `supported`, `unsupported`, and `ask_user`. Stop when `harm.score` is at or above 1.5. Proceed when the chosen label's probability is at or above 0.9 and `determined.noul` is at or above 0.9. Otherwise ask. Stop wins.

Proceed with `supported` means you may report the claim. It is not a substitute for the test you already ran. `unsupported` means say what is missing.

On `"source": "fallback"`, fail open. Report the receipts you have, or say you do not know. Do not invent a probability.

Template: `examples/verify.request.json`.


----- skills/jev-triage/SKILL.md -----

---
name: jev-triage
description: >
  Use when a support message, bug report, or other record needs one label
  such as billing, bug, or how-to, plus a harm score for a wrong label.
  Runs @aitofy/jev-awesome-skills. Do not send a mailbox or delete messages.
---

# Triage

Judge records you already have. Do not connect to a mailbox, delete, reply, or move messages as part of this skill.

Preset labels are `billing`, `bug`, `how_to`, `other`, and `ask_user`. Replace them with the labels you and the user agreed, and keep `ask_user`. One record per request. Put the record text in `state` and run:

```
jev questions triage
jev --dry-run --file request.json
```

Show the dry-run before a live call when the text is private. The package is `@aitofy/jev-awesome-skills`. Do not write a second client.

Stop when `harm.score` is at or above 1.5. Proceed when the chosen label's probability is at or above 0.9 and `determined.noul` is at or above 0.9. Otherwise ask. Stop wins. Proceed means apply the label in your tracker. Ask means leave it for a person.

On `"source": "fallback"`, fail open and leave the record unlabeled. Do not invent a probability.

Template: `examples/triage.request.json`.


----- skills/jev-review/SKILL.md -----

---
name: jev-review
description: >
  Use when a diff should be marked ship, fix, or ask a person, including when
  a change weakens a test or misses the task. Jev does not merge and does not
  replace CI. Runs @aitofy/jev-awesome-skills.
---

# Review

Read the diff and the test receipt yourself. Linters and tests stay in charge of syntax and exit codes. Jev only judges the closed question: ship, fix, or ask a person.

```
jev questions review
```

Put the goal, the relevant hunk, and what was actually run in `state`. Run `jev` from `@aitofy/jev-awesome-skills`. Do not write a second client. Do not merge, push, or dismiss a review because a probability is high.

Stop when `harm.score` is at or above 1.5. Proceed when the chosen label's probability is at or above 0.9 and `determined.noul` is at or above 0.9. Otherwise ask. Stop wins.

`ship` means you may say the change looks acceptable. `fix` means request a code change. Neither one is a green CI check.

On `"source": "fallback"`, fail open and review it yourself. Do not invent a probability.

Template: `examples/review.request.json`.


----- skills/jev-act/SKILL.md -----

---
name: jev-act
description: >
  Use when the next step must be one action from a list you already observed,
  such as a visible control or a legal game move. Jev cannot invent clicks,
  coordinates, or tools. Runs @aitofy/jev-awesome-skills.
---

# Act

Build the list before you call Jev. Every option must be an action you observed or a move your rules allow. Keep the presets `abstain` and `reobserve`.

```
jev questions act --option click_appearance="The Appearance row is visible in Settings"
```

Put the goal and the observation in `state`. Do not send passwords, cookies, payment data, or a raw screenshot. Run `jev` from `@aitofy/jev-awesome-skills`. Do not write a second client.

Stop when `harm.score` is at or above 1.5. Proceed when the chosen label's probability is at or above 0.9 and `determined.noul` is at or above 0.9. Otherwise ask. Stop wins.

Proceed means perform that one listed action, then observe the result yourself. `reobserve` means change nothing. `abstain` means stop and ask. Never derive an action that was not in the criteria.

On `"source": "fallback"`, fail open: reobserve or ask. Do not invent a probability or a click.

Template: `examples/act.request.json`.