@tony0721/pi-j-space

J-Space Cognition Suite for Pi — inference-time cognitive control layer for deep reasoning, long-horizon work, tool use, verification, and recovery. Ships the j-space skill plus a per-turn forcing extension.

Packages

Package details

extensionskill

Install @tony0721/pi-j-space from npm and Pi will load the resources declared by the package manifest.

npm repo home report

$ pi install npm:@tony0721/pi-j-space

Package: @tony0721/pi-j-space
Version: 1.1.0
Published: Aug 18, 2026
Downloads: 266/mo · 266/wk
Author: tony0721
License: unknown
Types: extension, skill
Size: 243.9 KB
Dependencies: 0 dependencies · 0 peers

Pi manifest JSON

{
  "extensions": [
    "./index.js"
  ],
  "skills": [
    "./skills"
  ]
}

Security note

Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.

README

j-space — Pi package

The J-Space Cognition Suite packaged for Pi as an official plugin: a skill + a forcing extension that keeps the operating protocol active in every conversation turn.

J-Space is an inference-time cognitive control layer built on the accessible representational space the model is poised to say. It manages what stays active, preserves constraints across long tasks, externalizes durable state, detects reasoning failure, and returns verified results in clean language. No weight changes, no fine-tuning, no hidden service.

The package is published on npm as @tony0721/pi-j-space (npm:@tony0721/pi-j-space, public). The full research suite lives at J-Space-Cognition-Suite-V3.6 (README, benchmarks, citation, license).

Install

pi install npm:@tony0721/pi-j-space

Restart Pi after installation. The package registers:

Skill skills/j-space — the complete suite: SKILL.md + modules/ (9 modules) + references/ + scripts/.
Extension index.js — keeps j-space active every turn.

How it works

Three independent injection layers (verified against the Pi main loop; default mode is compact so context stays lean):

Layer	Hook	Effect (≈tokens)
Per-message gate	`input`	Prepends a one-line gate header to every user message, unless it is a slash command (~40 tk)
System prompt	`before_agent_start`	Injects the compact protocol block into the system prompt per submission (~200 tk)
Payload fallback	`before_provider_request`	Ensures the `[[J-SPACE-ACTIVE]]` marker is present in the final LLM request payload

Injection modes (/jspace):

compact (default) — ~200-token protocol block; the model loads SKILL.md / modules on demand via read, following j-space's own selective-loading design.
full — injects the entire SKILL.md (~3.8k tk); use only when needed.
reminder — no system block; the per-message gate header carries the frame.

Usage

Commands must start with / (Pi command prefix):

/jspace                     Status (shown as a TUI notification)
/jspace on|off              Enable / disable forcing
/jspace compact|full|reminder   Switch injection mode

Runtime state persists in <agent dir>/j-space/state.json — ~/.pi/agent/ under Pi, ~/.prime/agent/ under Prime Agent (auto-detected; override with JSPACE_STATE_DIR).

Configuration

Env var	Purpose
`JSPACE_SKILL_PATH`	Override the skill `SKILL.md` path (default: the copy inside this package)
`JSPACE_STATE_DIR`	Override the runtime state dir (default: auto-detected `<agent dir>/j-space`, Pi → `~/.pi/agent`, Prime Agent → `~/.prime/agent`)
`JSPACE_DEBUG=1`	Write an `inject.log` under the runtime state dir

Development

npm pack --dry-run   # inspect the published tarball
node verify.mjs      # self-check: factory loads, skill/module paths resolve, payload escort works
pi -e ./index.js -p "hi"   # smoke-test the extension in print mode

License

Apache-2.0. See LICENSE and the suite repository for authors, citation, and provenance.