dsh-simple-plan
Manifest validDeepSeek Harness plugin that bundles the simple-plan skill: plan, execute with worker subagents, verify every deliverable with memoryless verifier subagents, and fix in at most three cycles. The Skill is shipped by the bundle, so there is nothing to copy into a skills directory.
simple-plan
Plan → Execute → Verify → Fix, enforced on every task, as a DeepSeek Harness plugin. The skill ships inside the bundle: install the plugin and the skill is there, with nothing to copy into a skills directory.
A minimal planning skill: break the request into 2–7 steps, execute each with a worker subagent, verify every deliverable with an independent memoryless verifier subagent, and fix in at most three cycles before reporting honestly.
Install
DeepSeek Harness Desktop — open Plugins in the sidebar, choose Add plugin, and enter:
https://github.com/bauerelizabeth07139/simple-plan
Then switch the new dsh-simple-plan bundle on. The Desktop app boots the
reserved desktop profile, so that is where it has to be enabled.
dsh CLI — install it into the profile you actually boot:
dsh plugin --profile web add bauerelizabeth07139/simple-plan
No git on the machine? pnpm resolves a git shorthand with git ls-remote,
which fails with 'git' is not recognized when git is missing. Use the tarball
instead — that path is plain HTTPS:
dsh plugin --profile web add https://codeload.github.com/bauerelizabeth07139/simple-plan/tar.gz/master
The same address works in the Desktop Add plugin dialog. Replace master
with a commit SHA to pin an exact revision (/tar.gz/<sha>).
Uninstall with dsh plugin --profile web remove dsh-simple-plan.
What it does
- Plan — understand the request, break it into 2–7 concrete steps, each with a deliverable.
- Execute — spawn a worker subagent per non-trivial step. A worker gets the task, the deliverable and the paths; it never gets the acceptance criteria.
- Verify — spawn a verifier subagent per deliverable. Verifiers are
memoryless (
subagent, neversubagent_fork) and report PASS or FAIL with evidence. - Fix loop — on FAIL, read the evidence, spawn a new worker for the specific defect, then a new verifier. At most three cycles.
- Report — summarise what was done, list each verification result, and state ALL PASS or FAIL.
The full instructions live in skills/simple-plan/SKILL.md.
The plugin registers that file as a bundled skill through the Harness skill
registry, so it can be triggered by the model and invoked by the user.
How it differs from the Codex original
The rules are unchanged; the tool names are the Harness's:
| Codex | DeepSeek Harness |
|---|---|
multi_agent_v1__spawn_agent(message=…, fork_context=false) | subagent(description=…, prompt=…) |
multi_agent_v1__wait_agent(targets=[…]) | subagents report back on their own; the parent collects results |
apply_patch | edit / write |
shell | pwsh (Windows) or the session's shell tool |
read_file | read |
The one substantive gain: DSH gives every agent read_image, so a verifier can
inspect a screenshot even when the session model has no built-in vision. The
skill tells it to.
Development
No build step and no runtime dependencies — @deepseek-ai/cordis and
@deepseek-ai/dsh-skill are peers supplied by the Harness.
npm test # node >= 22: manifest, card metadata, loader patch, skill, syntax
index.js is the whole host half: it registers the bundled skill provider.
cordis.patch.yml is the loader row that mounts it.
Repository layout
| Path | Purpose |
|---|---|
index.js | the DSH plugin: registers the bundled skill |
cordis.patch.yml | the loader row that activates the plugin |
skills/simple-plan/SKILL.md | the skill itself, in Harness tool names |
locale/{en,zh}.json | card title and description for the plugin lists |
assets/icon.svg | card artwork |
test/plugin.test.mjs | static composition checks |
SKILL.md, AGENTS.md | the original Codex skill and instructions, unchanged |
Codex usage (unchanged)
The repository still contains the original files, byte for byte. On Codex:
git clone https://github.com/bauerelizabeth07139/simple-plan.git
cd simple-plan
cp -r . ~/.codex/skills/.system/simple-plan
No AGENTS.global.md is needed for the Codex skill. Note that DSH also reads
AGENTS.md — including a user-global $DSH_HOME/AGENTS.md — so if you want the
rules applied outside a skill invocation, that file is the place.
License
Comments
Loading…
Similar plugins
by bauerelizabeth07139
DeepSeek Harness plugin that bundles the plan-guardian skill: a mandatory 7-step plan with binary acceptance criteria, execution through worker subagents, and completion validated by independent memoryless verifier subagents that report PASS or FAIL with evidence.
★ 0
MIT
JavaScript
Oct 4, 2026
dsh plugin --profile web add dsh-plan-guardianby KratosLee-6
DeepSeek Harness Plugins · A local-first DeepSeek Harness plugin for discovering, combining, and turning skills into actionable ideas.
★ 1
MIT
TypeScript
Nov 8, 2026
dsh plugin --profile web add dsh-plugin-sparkby abbaababa
自动任务分解 + 真实部门子 Agent + 「任务进度」面板,给 DeepSeek Harness 用;没有任何第三方运行时依赖。
★ 3
MIT
TypeScript
Nov 20, 2026
dsh plugin --profile web add dsh-workplanby 82c86b8z86-stack
Engineering workflow layer for DeepSeek Harness (dsh): a disciplined-engineer agent preset with five gated phases — requirements clarification, plan approval, TDD, parallel subagent execution, and ver
★ 4
MIT
JavaScript
Oct 11, 2026
dsh plugin --profile web add dsh-engineering-workflowby runcat-tommy
Minimal skill manager for DeepSeek Harness Web: a Settings page beside Plugins that lists every installed skill with its own summary, source and disk footprint, and removes any of them into a recovera
★ 0
MIT
JavaScript
Oct 16, 2026
dsh plugin --profile web add dsh-skill-mini-managerby hasan-aghayev
Automatically orchestrates complex DeepSeek Harness requests. It detects when delegation is useful, builds a dependency-aware plan, assigns bounded researcher, architect, backend, frontend, tester, an
★ 0
MIT
TypeScript
Oct 22, 2026
dsh plugin --profile web add dsh-task-orchestrator