redreamality/jev-skill-selection
User repository on GitHub
Pre-message hook: use TypeSafe Jev to keep/drop skills and shrink agent context
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Agents & MCP
Jev Skill Selection
A Python keep/drop gate that decides which SKILL.md files enter the prompt before the first model message, from offline keyword overlap or from Jev Noul and Choice scores.
- Stars
- 2
- Language
- Python
- License
- MIT
- Published by
- Community
- Author
- redreamality
- Last verified
- Oct 2026
About this project
Stop dumping every SKILL.md into the prompt. Keep/drop skills before the first model message — shrink context, save tokens. Skill folders grow. Agents dump many SKILL.md files every turn. Unrelated skills burn tokens before the model even starts. Load a closed catalog of skills. selectskills(mode=local|jev) → SelectionResult {kept, dropped, charssaved}. Hook injects only kept skills into the prompt. Multi-skill keep/drop for context size — not a single-skill suggester.
Quoted from the project's README. Read the full README.
- Languages
- Python
Highlights
- Standard library only, Python 3.10+, and local mode needs no key: the demo keeps one of five fixture skills and drops four.
- Hard filtering keeps dropped skills out of the tool and skill index entirely; soft inject stays available via JEV_FILTER_MODE.
- select_skills returns kept, dropped and chars_saved, and the pre-message hook injects only the kept blocks.
- Adapters for Claude Code, Codex, Hermes and OpenCode each ship a hard path and an optional soft path.
Watch out
The default CLI threshold is 0.45 while the harness and live end-to-end runs use 0.1, and the README says to calibrate per catalog because local Jaccard overlap and Jev Noul are not directly comparable. The model is pinned to jev-1.13.0 for reproducibility, so the pin needs re-checking after releases. The savings figures come from five fixture skills and one short message, not from a large catalog. Hard filtering also depends on the host exposing it, and the Codex adapter may need a restart. Copy reviewed against the repository README on 2026-10-04, not from running it.
Questions about this entry
Short answers drawn from the linked project or page.
Different output and different timing. The cookbook suggests at most one skill name as a recommendation UX. This keeps or drops many skills and runs before the first model message, because its goal is context volume and token savings rather than a single-skill recommendation. It is explicitly not a single-skill suggester.
That is what the threshold is for, and the default is conservative deliberately: 0.45 in the CLI against 0.1 in the harness and live runs, with an explicit instruction to calibrate per catalog. The default noul primitive keeps several skills rather than picking one, and choice mode is available as a soft ranking for small catalogs. Local overlap being too blunt for your catalog is the stated reason to switch to Jev mode.
It saves them where the host supports it, because dropped skills never enter the tool or skill index and cannot be read into the prompt at all. Soft inject, which most adapters ship by default, only tells the model which skills to prefer and still pays the context cost. The adapters under adapters/ default to hard, with soft or both selectable through JEV_FILTER_MODE.
Not in local mode, which is the default and does keyword and metadata shortlisting entirely offline with no key and no third-party runtime dependencies. Jev mode posts to the TypeSafe systemone endpoint with TYPESAFE_API_KEY and pins jev-1.13.0. A live end-to-end script with a mock LLM, plus optional real Jev, is included.
Python 3.10+ and pip install -e ., then python -m jev_skill_selection select with --mode local runs the five-fixture demo with no key: on the sample message it keeps git-ops and drops four skills, cutting 1,431 characters to 393.
Claude Code, Codex, Hermes and OpenCode, each with a hard path such as skillOverrides set to off, skills.config enabled false (which may need a restart), stripping available_skills, or filtering the tool definition. JEV_FILTER_MODE switches it to soft or both.
Yes. load_catalog and select_skills with SelectionOptions return kept, dropped and chars_saved, and the host-agnostic before_first_message hook returns prompt blocks for only the kept skills.
More Jev resources like this
51GitHub stars
Local LinkedIn recruiting workspace on Browser Use's Jev Ultrafast: Jev screens titles, drives navigation, and assesses profile excerpts against your brief while a live browser stays visible; matches and evidence export as JSON.
Jev Agents & MCP#community#jevbest#catalog-import
Community
skeptrunedev/jev-recruiterPython
skeptrunedev/jev-recruiterPython33GitHub stars
Codex Skill plus stdlib-only Python CLI that recommends which installed skill to read next for a task, allowing no-skill or uncertain answers and never executing candidates.
Jev Agents & MCP#community#jevbest#catalog-import
Community
win4r/jev-skill-suggesterPython
win4r/jev-skill-suggesterPython72GitHub stars
MCP server over stdio that gives coding agents fourteen Jev judgment tools, turning evidence plus an enumerable question into auto, review, or escalate actions from calibrated probabilities.
Jev Agents & MCP#community#jevbest#catalog-import
Community
PyModel/jev-judge-mcpPython
PyModel/jev-judge-mcpPython21GitHub stars
Research runtime that pages a large logical tool and context space through a small bounded Jev decision surface with explicit approvals.
Jev Agents & MCP#community#jevbest#catalog-import
25GitHub stars
Agent-oriented playbook for using Jev as a cheap, fast decision layer that triages, classifies, scores, and gates work before expensive browser, research, retry, or subagent steps.
Jev Agents & MCP#community#awesomejev#catalog-import
Community
Bodila51/muse-jev-playbookPython
Bodila51/muse-jev-playbookPython14GitHub stars
Opt-in Hermes plugin that routes each turn to one skill: a wide Jev Choice plus request checks gate a narrow fits round, injecting a single ignorable hint line.
Jev Agents & MCP#community#jevbest#catalog-import
From the Jev community
Posts from builders shipping with Jev right now.