safzanpirani/pi-jev-skill-picker
User repository on GitHub
Rank Pi Agent Skills for the current task with TypeSafe Jev
- Forks
- 6
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Agents & MCP
Pi Jev Skill Picker
Pi extension that strips the full Agent Skills catalog from every request and loads only the skills a Jev ranking call judges relevant.
- Stars
- 36
- Language
- TypeScript
- License
- MIT
- Published by
- Community
- Author
- safzanpirani
- Last verified
- Oct 2026
About this project
A Pi extension that keeps the Agent Skills catalog out of model requests and replaces it with one ranking tool backed by TypeSafe's Jev System One model. Skills stay loaded, so /skill:name keeps working. Before each agent turn the extension removes Pi's generated catalog from the effective system prompt and gives the model two tools to pull skills in on demand.
Quoted from the project's README. Read the full README.
Used inJev for Software & DevTools
- Languages
- TypeScript
Highlights
- Removes the generated skills block from the system prompt and adds a skill_search tool backed by the Jev decision model.
- Each skill is scored independently on unrelated, adjacent, and directly-applicable levels with a configurable floor.
- skill_load loads one to five named skills off disk with fuzzy matching and suggestions for bad names.
- Configurable model, shard size, score floor, skill count, timeout, endpoint, and description truncation.
Watch out
Needs a TypeSafe-compatible key and endpoint configured; ranking quality depends on skill descriptions and threshold tuning. 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.
Direct naming still works because skills stay loaded. skill_search ranks by task and loads winners in full, while skill_load fetches a known name straight off disk with no model call.
It sends one task string and gets back scored skills, instead of forking the whole conversation into a child agent with a persisted session like the older subagent picker did.
The example 137-skill catalog ran about 19,000 tokens per request, while one ranking call costs a fraction of what a single unstripped request wastes against the listed models.
It falls back to deterministic lexical matching and says so in the result, returning metadata and paths so the agent decides what to read.
Each skill becomes its own Score question with the task in the shared state and the skill's name and description in that skill's instructions, so no keyword prefilter can drop one first. Every skill is rated on three ordered levels: 0 unrelated, 1 adjacent, 2 directly applicable. Code applies the minScore floor, sorts, and loads the winners, with ties breaking on confidence.
Environment variable, then skill-jev.json under PI_CODING_AGENT_DIR (normally ~/.pi/agent), then the package default. Knobs include the API key, model (jev-latest), shardSize (50), minScore (1.6), maxSkills (3), timeoutMs (20000), descriptionLimit (1200), and endpoint. Raise minScore if too many adjacent skills load, lower it if a relevant skill is missed.
No. Both tools return skill content as tool results and neither writes to the system prompt, which is what keeps the cached prefix stable across turns. The extension strips Pi's generated skills catalog from the effective system prompt before each turn, while skills stay loaded so /skill:name keeps working.
More Jev resources like this
102GitHub stars
Behavioral code search that asks Jev to score repository fragments for a plain-language question and returns exact excerpts with paths and lines.
Jev Agents & MCP#community#jevbest#catalog-import
Community
nassim-arifette/jevgrepTypeScript
nassim-arifette/jevgrepTypeScript7GitHub stars
MCP server exposing Jev Choice, Score, Noul, batched ask, file-reading triage, and model listing as typed tools with full probability distributions.
Jev Agents & MCP#community#jevbest#catalog-import
Community
rashedInt32/jev-mcpTypeScript
rashedInt32/jev-mcpTypeScript5GitHub stars
A Claude Code function hook that sends AskUserQuestion dialogs to Jev and draws a probability per option above the dialog, so you still choose while seeing how confident the model is.
Jev Agents & MCP#community#jevbest#catalog-import
Community
BeLazy167/typesafe-modTypeScript
BeLazy167/typesafe-modTypeScript3GitHub stars
An MCP server built around measuring questions rather than asking them: jev_ask prototypes wording, jev_eval sweeps thresholds against labeled data, jev_map applies the winner over a workload.
Jev Agents & MCP#community#jevbest#catalog-import
Community
BYK/jev-mcpTypeScript
BYK/jev-mcpTypeScript0GitHub stars
A one-commit TypeScript MCP server that exposes three TypeSafe System One judgments — noul, choice and score — as agent tools, with src, tests, docs, scripts and a knip config. No README, zero stars, no topics.
Jev Agents & MCP#community#jevbest#catalog-import
0GitHub stars
Local stdio MCP server that lets an agent delegate a decision: send a decision, context and 2 to 64 choices, get back a recommended choice with per-choice percentages and an explicit statement of where those percentages came from.
Jev Agents & MCP#community#jevbest#catalog-import
Community
dakdevs/decide-mcpTypeScript
dakdevs/decide-mcpTypeScriptFrom the Jev community
Posts from builders shipping with Jev right now.