DeepSeek Harness plugin

deepseek-desk-rsi

Recursive Self-Improvement (RSI) engine for DeepSeek Harness: a bounded perceive → integrate → verify → repair → propose loop that keeps a downstream desktop app in sync with the fast-moving upstream

Jump to install

Source facts

Repository
huchunlinnk/deepseek-desk-rsi
Latest update
Aug 18, 2026
Category
Development & Runtime
GitHub stars
1
Format
plugin
Catalog evidence
Upstream dsh.bundle evidence
Evidence path
package.json#dsh.bundle
Checked against
0.1.0-rc.8
Upstream check date
2026-08-20

This evidence comes from the upstream catalog. This site has not installed, run, or security-reviewed the plugin.

Install

Start with a prompt that asks an agent to review the GitHub repository and source. Switch to the command if you want to install it yourself.

Copy this prompt into DSH, Codex, or another agent and ask it to review the GitHub repository and source first.

Do not install or run any commands yet. Read this plugin's GitHub repository, README, and relevant source code. Then answer the questions below clearly and directly so I can decide whether it fits my needs:

1. What is this plugin, and what problem does it solve?
2. Who is it for, and what are its typical use cases?
3. How is it used after installation? Include one minimal example.
4. What known limitations or privacy, security, compatibility, or maintenance risks does it have?
5. Give a clear recommendation: recommend, conditionally recommend, or do not recommend, with reasons.

Distinguish statements documented by the repository, inferences from source code, and unknowns. If evidence is insufficient, say so explicitly. Do not guess or simply repeat the README.

GitHub: https://github.com/huchunlinnk/deepseek-desk-rsi
Plugin: deepseek-desk-rsi
Author: huchunlinnk

Check the source files

Read the README and other files from this plugin directory before installing.

File explorer4 files
README.en.mdSource · read only
README language

deepseek-desk-rsi

English | 中文

![License: MIT](./LICENSE)

A DeepSeek Harness (DSH) plugin that makes a downstream app keep itself in sync with the fast-moving upstream harness — using DSH itself as the worker.

AI for AI: DSH maintains DSH.

Why this exists

DeepSeek Harness is in 0.1.x-rc and ships breaking changes often. A desktop shell (or any downstream app) that pins one version rots within days. Chasing upstream by hand is a full-time job. This bundle hands that job to an agent — with hard, automatable success signals so the loop can't drift.

Why — the biggest advantages

1. AI for AI, genuinely. The worker is DeepSeek Harness itself: it diffs upstream, edits the integration glue, builds, tests, rolls back, and opens a PR — no human in the loop except the final merge. 2. "一模一样" is machine-checked, not a promise. rsi_parity enforces a 128-plugin contract (parity.json, generated from the official dsh-base + dsh-web-app bundles). Any upstream add/rename/remove surfaces as a hard failure — exact-name matching, not substring. 3. Evolution that cannot run away. Bounded repair (max 3 rounds), Memento checkpoints (rsi_checkpoint / rsi_rollback), and a PR gate — the loop proposes, a human merges. 4. Zero-build install. Plain ESM + JSDoc, no prepare script: dsh plugin add github:huchunlinnk/deepseek-desk-rsi#main just works.

The loop

perceive  →  integrate  →  verify  →  parity  →  repair (bounded)  →  propose
StageWhoHow
perceivetool rsi_perceivefetch upstream, diff commit log + stat since last sync
integrateagent (existing tools)apply the relevant harness changes to the app's glue with bash / edit / subagent / workflow
verifytool rsi_verifythe hard fitness function: cargo build && cargo test && bash scripts/smoke-web.sh, ok only on exit 0
paritytool rsi_paritythe identical-surface gate: every required plugin in parity.json is still present
repairagent + rsi_checkpoint / rsi_rollbackcheckpoint before edits (Memento), roll back and retry after a failed verify/parity
proposetool rsi_proposecommit, push a branch, open a PR via gh — the human gate

The loop discipline is enforced by a system-prompt section (rsi:loop) that the bundle injects, so the agent cannot silently skip a stage.

Tools

ToolPurposeDesign pattern
rsi_perceivediff upstream since a base ref
rsi_checkpointcommit a labeled snapshotMemento (capture)
rsi_rollbackgit reset --hard back to a checkpointMemento (restore)
rsi_verifyrun the build/test/serve fitness command
rsi_paritycheck parity.json against --dump-config (exact-set)
rsi_proposeopen a PR via ghCommand (review gate)

The five stages are a Pipeline; the whole loop is bounded recursion (max 3 repair rounds).

Install

dsh plugin --profile rsi add github:huchunlinnk/deepseek-desk-rsi#main
dsh --profile rsi "Run the daily upstream sync: perceive, integrate, verify, parity, repair if needed, then propose a PR."

The profile/ directory is a starting headless profile (dsh-base + dsh-headless + dsh-desk-rsi). Copy it into $DSH_HOME/profiles/rsi and point it at your repo.

Parity contract

parity.json lists every plugin in the official dsh-base + dsh-web-app bundles (128 plugins) plus the host endpoint. rsi_parity fails if any name goes missing after a sync. Regenerate it from the official bundles whenever upstream adds/renames a plugin:

node scripts/gen-parity.mjs /path/to/deepseek-harness

Design principles

  • First principles — reuses the harness (@deepseek-ai/dsh-tools, @deepseek-ai/cordis are peer deps provided by DSH); never reimplements bash, git, or subagents.
  • Occam's razor — plain ESM JavaScript, zero build, zero prepare; a git install works out of the box.
  • Design patterns — Pipeline (stages), Memento (checkpoint/rollback), Command (PR gate), Strategy (the desktop's update policy), bounded recursion.

Repository layout

index.js           single entry; registers the 6 tools + the loop prompt
prompt.js          the rsi:loop system-prompt section (discipline)
tools/             one tool per file
lib/exec.js        bounded shell runner
scripts/           smoke.js, e2e-loop.mjs, gen-parity.mjs
profile/           starting headless profile
parity.json        128-plugin contract

Testing

npm install --no-save --no-package-lock @deepseek-ai/dsh-tools   # prebuilt registry + cordis peer
npm test   # smoke (register 6 tools + prompt) + e2e (real-git perceive→checkpoint→rollback→verify→parity)

To develop against a locally built harness checkout instead, symlink its packages into node_modules/@deepseek-ai/ (as the CI once did).

Security

A running loop is a local user with shell access. Secrets never enter source; the PR is the gate; pin upstream to a commit you trust. See [SECURITY.md](./SECURITY.md).

License

[MIT](./LICENSE)