J-Space Cognition Suite V3.7
J-Space Cognition Suite is a model-agnostic inference-time control system for deep reasoning, long-horizon work, tool use, verification, and recovery.
It is packaged as a Skill for cross-platform use, selective loading, and low-friction integration.
The suite organizes an agent's accessible working representations into a deliberately managed workspace. It operates through a single entry, nine selectively loaded modules, four supporting references, and an optional standard-library controller for durable task state.
J-Space operates at inference time. Model weights and training remain unchanged.
Quick start
Option A — manual installation
-
Download or clone this repository.
-
Locate the user-level Skills directory used by your AI host.
-
Copy the complete
j-space/directory into it so that the installed entry is<skills-directory>/j-space/SKILL.md. -
Run the integrity check with an available Python 3 interpreter:
<python-command> <skills-directory>/j-space/scripts/verify_suite.pyReplace
<python-command>with the Python 3 command available on the host, commonlypython,python3, orpy -3. -
Reload the host if it discovers Skills at startup.
-
The directory must remain intact because
SKILL.mdroutes to relative paths undermodules/,references/, andscripts/. -
The repository-level
LICENSEandTHIRD_PARTY_NOTICES.mdremain part of the distribution. -
Include copies of both when redistributing
j-space/as a standalone package.
Option B — ask an AI agent to install it
Copy the following prompt into an agent that can access files and this repository:
Install J-Space Cognition Suite from
https://github.com/Tiger3807861189/J-Space-Cognition-Suite-V3.7 into this environment's user-level Skills directory.
First inspect the host configuration or documentation to locate the correct Skills directory. Install the complete j-space/ directory as j-space/, preserving SKILL.md, modules/, references/, and scripts/. If a j-space target already exists, compare it and ask before replacing anything. Run scripts/verify_suite.py with an available Python 3 interpreter after installation.
When finished, report the installed path and verification result, then tell me how this host invokes the Skill. Briefly explain fast, full, and loop, and explain that the optional controller records long-task state rather than choosing solutions. If this host has no native Skill loader, explain the selective system/developer-instruction integration instead of reporting an installation.
Use it
Invoke the Skill through the mechanism provided by your host—such as its Skill picker,
/j-space, $j-space, or a direct request:
Use j-space for this task. Audit this repository, preserve its architecture,verify every finding, and keep the work consistent across all affected files.
The entry gate selects the lightest suitable pass automatically.
Operating modes
| Pass | Suitable work | What loads |
|---|---|---|
fast |
One step, or a result checkable in one glance | Nothing extra |
full |
Several dependent steps and one bounded deliverable | One or two relevant modules; ship before delivery |
loop |
Multiple stages, files, turns, tools, or persistent state | Ledger, seams, checkpoints, register audit, and recovery |
A request for brevity changes the outer response length while verification remains aligned with the task's floor. Short work stays light; long work receives durable state only when it needs it.
Core mechanisms
| Mechanism | Function |
|---|---|
| Selective workspace loading | Keeps one or two load-bearing ideas active and externalizes the rest |
| Broadcast hub | Gives dependent branches one shared source for names, values, constraints, and style anchors |
| Dense Track | Carries long internal chains in compact, decodable notation before returning to clean outer language |
| Bridge-before-conclusion reasoning | Makes required intermediates explicit before a conclusion consumes them |
| Metacognitive control | Routes confidence, inconsistency, and failure signals into a concrete next action |
| Empirical escape and verification | Converts stalled derivation into bounded tests with a named verifier and coverage |
| First-person agency and functional echo | Uses I, we, let's, and we need to bind workspace state to later actions and checks |
The mechanisms are selectively loaded. They are not a fixed checklist for every request.
Optional controller
j-space/scripts/jspace.py externalizes loop state into
.jspace/ in the current task workspace. Invoke it by its resolved Skill path while keeping the task workspace as the current directory.
| Command | Purpose |
|---|---|
note --goal "..." --next "..." |
Open the ledger and define done plus the first action |
note --next "..." |
Replace the single next action after a checkpoint or seam |
note --core "name — defining fact" |
Record a hub entry |
note --core "name — defining fact" --core-slot 1 |
Swap a selected live hub entry |
note --check "..." --by "..." |
Append a checkpoint with verifier and coverage |
note --open "..." --settled-by "..." |
Record a question and what would settle it |
note --close N --check "..." --by "..." |
Close question N against a new recorded checkpoint |
seam |
Re-read current state and report recent movement |
ship FILE |
Inspect outgoing text for register leakage and failure signatures |
resume |
Reload the premise, invariants, and full ledger after a long gap |
<python-command> <skill-root>/scripts/jspace.py note --goal "what done means" --next "first action"
<python-command> <skill-root>/scripts/jspace.py note --open "does the parser preserve state?" --settled-by "unit tests over all ledger sections and edge inputs"
<python-command> <skill-root>/scripts/jspace.py note --close 1 --check "the parser preserves state" --by "unit tests over all ledger sections and edge inputs"
<python-command> <skill-root>/scripts/jspace.py seam
<python-command> <skill-root>/scripts/jspace.py ship OUTPUT_FILE
<python-command> <skill-root>/scripts/jspace.py resume
The controller records and reports state. Solution choice remains with the model. It uses the Python standard library and writes working state only under the task's .jspace/ directory.
Generic model integration
An environment with a native Skill loader can install j-space/ directly. For a chat or API environment, provide j-space/SKILL.md as a system- or developer-level instruction and expose modules/ and references/ through file or retrieval tools.
Selected files should be retrieved on demand. Selective loading is part of the operating design.
Benchmark
1. Main table
| Benchmark | DeepSeek V4-Flash-Vision-Exp | DeepSeek V4-Flash-Vision-Exp + J-Space V3.7 | GLM-5.3 | Opus-4.8 | Fable 5 (w/ fallback) |
|---|---|---|---|---|---|
| HLE (w/o tools) | *37.8 | 37.8 | — | 49.8 | 53.3 |
| HLE (w/ tools) | *51.5 | 51.9 | 62.5 | 57.9 | 63.0 |
| Terminal Bench 2.1 | 83.9 | 85.5 | 88.2 | 85.0 | 88.0 |
| NL2Repo | 57.7 | 60.4 | 58.0 | 69.7 | — |
| CyberGym | 75.3 | 77.8 | 84.5 | 78.3 | 83.1 |
| DeepSWE | 59.3 | 61.8 | 66.9 | 58.0 | 70.0 |
| Toolathlon-Verified | 75.9 | 77.4 | 73.0 | 76.2 | 77.9 |
| Agents' Last Exam | 27.3 | 28.3 | 28.5 | 25.7 | 23.8 |
| AutomationBench (Public) | 25.7 | 27.6 | 48.2 | 27.2 | 29.1 |
| *Average | 56.99 | 58.61 | 64.54 | 58.33 | 62.13 |
* HLE scores were not disclosed and follow DeepSeek V4-Flash-0731. The average covers the 7 rows where all 5 columns have values.
2. Speed and token efficiency
| Benchmark | Wall-clock τ | Speedup | Output tokens | Total tokens | Accuracy multiplier | Score per unit time | Cost per successful task |
|---|---|---|---|---|---|---|---|
| HLE (w/o tools) | *1.02 | −2% | −10% | +5% | 1.000 | 0.98× | +5% |
| HLE (w/ tools) | 0.88 | +14% | −22% | +3% | 1.008 | 1.15× | +2% |
| Terminal Bench 2.1 | 0.79 | +27% | −28% | −3% | 1.019 | 1.29× | −5% |
| NL2Repo | 0.76 | +32% | −31% | −5% | 1.047 | 1.38× | −9% |
| CyberGym | 0.78 | +28% | −28% | −2% | 1.033 | 1.32× | −5% |
| DeepSWE | 0.78 | +28% | −28% | −3 |