dsh-j-space
Bundles the J-Space Cognition Suite V3.6
as a DeepSeek Harness plugin skill: a model-agnostic, inference-time "inner
workspace" operating protocol (reasoning discipline, planning, long-horizon
state, verification, and a durable task ledger). No model weights are touched;
nothing runs unless the agent invokes the skill.
What it does
- Registers a bundled skill provider into the host skill registry (global
layer), so every agent preset catalog gains the
j-space skill.
- Serves the full suite — entry skill, nine modules, references, and the
optional
jspace.py ledger controller — from inside the package.
- Injects a small runtime note resolving
<skill-root> and the Python
command, matching the suite own portability instructions.
- Registers nothing else: no tools, no prompts, no UI. If the host has no
skill registry, the plugin is a no-op and startup is unaffected.
Install
In a DSH profile (~/.dsh/profiles/<name>):
dsh plugin --profile <name> add ./artifacts/dsh-j-space-0.1.0.tgz
# or directly from a GitHub release:
dsh plugin --profile <name> add https://github.com/Harzva/dsh-j-space/releases/latest/download/dsh-j-space-0.1.0.tgz
Then restart the profile. In a session, the agent (or you) can invoke it:
use skill j-space
The skill writes its .jspace/ ledger into the current task workspace, so it
stays with the task. It needs only a Python 3 interpreter (python3) for the
optional ledger controller; every ledger behavior has a hand-executable
equivalent in the modules, so nothing is lost without it.
Development
pnpm check # node --check + unit tests
pnpm run pack:dsh # builds artifacts/dsh-j-space-<version>.tgz
pnpm run verify:dsh-offline # isolated DSH profile boot + registry integration
License
Plugin code: Apache-2.0. Vendored suite content: Apache-2.0, © its authors —
see THIRD_PARTY_NOTICES.md and the vendored upstream provenance in
UPSTREAM_COMMIT.
