Jev Skill Gate — Development skill for Claude Code
Cut Claude Code's skill manifest by ~75% with TypeSafe Jev.
How to install Jev Skill Gate
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open ShivamPansuriya/jev-skill-gate and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Jev Skill Gate does
Cut Claude Code's skill manifest by ~75% with TypeSafe Jev. Scores every installed skill for relevance and hides the rest via skillOverrides — 12,750 → 3,185 tokens on a 217-skill install, for $0.0009 a session.
Alternatives in Development
- Sentiment Analysis — Analyze user feedback with sentiment scores and JTBD insights 7.8k ★
- Verify Theme — Run the full Graphite quality gate (drift, tokens, contrast, YAML) and report pass/fail 452 ★
- Voice Guard — Use to check ONE already-written piece of copy against a voice guide before it ships — it spawns ONE fresh sib 130 ★
README
jev-skill-gate
[](https://github.com/ShivamPansuriya/jev-skill-gate/releases) [](test/run.mjs) [](https://nodejs.org) [](package.json) [](LICENSE)
Claude Code loads every skill description into context at session start, whether or not the skill has anything to do with what you are working on. On a machine with a full skills library that is roughly 10,000 tokens spent before you type anything.
`jev-skill-gate` scores each skill against the project you are actually in, then writes `skillOverrides` so only relevant skills reach the model. Scoring runs on [TypeSafe's Jev](https://typesafe.ai), a decision model that returns calibrated probabilities instead of text, in one parallel pass over all your skills.
Does it pick the right skills?
That is the only question that matters — a gate that saves tokens by hiding what you needed is worse than no gate. So there is a committed eval, not a claim.
**20 labeled cases · 217 installed skills · every label validated against the live inventory before the run.** Each case lists skills that *should* rank high and skills that are unambiguously irrelevant. State is the **prompt only**, with no project signals helping, which is the harder test.
| Metric | Jev (4 cases) | Local scorer (18 cases) |
|---|---|---|
| Mean pairwise AUC | 1.000 | 0.961 |
| Cases with perfect separation | 4 / 4 | 14 / 18 |
| Median rank of an expected skill | 3 of 217 | 4 of 217 |
| Irrelevant skills reaching any top 10 | 0 | 0 |
| Expected skills surviving the gate | 100% (19/19) | 92.9% |
| Skills hidd |
Related Skills
Expired
Find expired domains with backlink equity in your niche. Scores each by DR, referring domains, topical relevan
Hammer
GATE H — must-have census, baseline comparison, cut list + ship verdict
Slm Gate
Decoupled local-SLM pre-processing and routing layer for MCP and cloud LLMs. Intercepts, distills, and routes
Claude Code Optimizer
Optimize Claude Code token usage — 22 skills, 8 hooks, 6 rules. One npm install. Zero manual steps. Cut costs
Build Install
Build the TangoDisplay macOS app and verify it is installed correctly.
Cc Mem
Query and manage cc-memory state for the current project. Subcommands run the installed plugin's CLI (cc_memor
Related Agents
Framework Discovery Agent
Framework discovery specialist. Given a topic and scope (product internal both), scans ADD artefact filenames
Source Librarian
Scores sources by reliability, freshness, relevance, and conflict risk. Builds the citation map. Flags claims
10x
10x is a terminal-based CLI coding agent (installed via npm install -g 10x-cli) built around the observation t
