btspoony/mstar-harness
Skill-driven harness/loop engineering workflow agent plugin.
Install
via GitHub · GitHub source
Installing from GitHub runs the project's build script, which pnpm blocks until you allowlist it — run the command once and pnpm prints the exact key to add under `allowBuilds` in ~/.dsh/profiles/web/pnpm-workspace.yaml.
What happened when we ran it
Installed cleanly when we ran it
Every command here is run in a throwaway container against a clean profile, and the result is whatever the harness recorded — not a guess from the source. Last run 13d ago.
Show it in your README
For maintainers: the badge serves this listing's latest sandbox verdict, so a re-run updates it on its own — and it links readers to the full result here.
Due diligence
What it is — It is a TypeScript Harness Workflow Engine plugin that executes deterministic workflow gates and drives multi-agent judgment through mstar-* skills.
Who it is for — It fits when you need to manage engineering workflows in DeepSeek Harness using the TypeScript engine for deterministic gates and multi-agent judgment. If you only need a non-TypeScript based simple plugin, you do not need to install it.
Watch out — Source installation from GitHub requires manually allowing the build script to run. It was discovered to execute shell commands during static checks. Sandbox test passed: it registers into the profile in a new setup. No obvious issues found.
The verdict — I would install it because it supports the same TypeScript engine and skills across multiple hosts.
Generated by grok-4.6, and a starting point rather than a verdict. Where it says a plugin installs or does not, that is from a real run in a clean profile — everything else is read off the repository. Trust the source over this.
mstar-harness is a DeepSeek Harness plugin that provides a skill-driven workflow engine for planning, implementing and checking code changes. DeepSeek Harness uses the plugin’s Agent Plugins v1 package, including plugin.json and skills/, while the TypeScript @mstar-harness/engine applies workflow gates for paths, status, leases, dispatch, SDD, iteration and linting. The mstar-* skills supply role-specific judgement and workflow decisions rather than leaving those rules solely in prompts.
The package also includes @mstar-harness/cli, which provides the mstar-harness and mstar commands for engine checks and host setup. Typical flows are a per-plan Prepare → Execute → QC → QA gate → Done cycle, an iteration running through Phases 1–5, or the read-only /codebase-audit flow, which writes plans under {PLAN_DIR}/audit-<date>/. Setting enforcement: hard in an iteration compass makes dispatch preflights fail fast; otherwise, checks remain advisory.
mstar-harness is intended for teams that want a prescribed, multi-agent delivery workflow across DeepSeek Harness and other supported hosts. It is a poor fit for lightweight sessions that do not need planning or gate enforcement, or for environments that cannot use the package’s terminal-facing CLI. Installation and bootstrap can run scripts and independent plugin installs, and the workflow can invoke terminal commands in the agent’s available permissions. The short mstar executable also conflicts with an unrelated npm package of the same name.
DeepSeek Harness loads the portable Agent Plugins v1.0.0 package from the repository root through plugin.json and skills/. The dsh setup described by the README uses the @mstar-harness/dsh adapter and can also install dsh-llm-fallbacks independently. The optional engine CLI is installed globally as @mstar-harness/cli; its canonical executable is mstar-harness, with mstar as a short alias.
The documented workflow setting is:
| Key | Effect |
|---|---|
enforcement: hard |
Makes dispatch preflights fail fast instead of leaving engine checks advisory |
This setting belongs in an iteration compass. The README does not specify a default value for enforcement.
The CLI exposes workflow validation commands cited by the skills, including mstar status validate, mstar dispatch validate and mstar iteration gate. Use mstar-harness instead of mstar when the shared executable name is ambiguous. npx @mstar-harness/cli doctor --target <opencode|cursor|codex|zcode|omp|dsh> checks a host setup, while npx @mstar-harness/cli plugin validate validates the portable package.
Workflow entry points include /iteration-start [direction] [pause], /iteration-drive, /iteration-loop [direction] [scale], and /codebase-audit [keywords]. scale accepts S, M, L or XL, defaulting to M; audit effort accepts quick or deep, with standard as the default. Audit output is read-only and goes to {PLAN_DIR}/audit-<date>/.
DeepSeek Harness can enter the project-manager workflow through the pm skill, which is not auto-loaded on dsh. The iteration flow covers Phase 1 through Phase 5, either interactively or autonomously. The engine enforces deterministic gates, while mstar-* skills remain the source of role and workflow judgement. Supported host adapters listed by the README are dsh, omp, OpenCode, Cursor, Kimi Code, ZCode and Codex.
The engine checks require @mstar-harness/cli on PATH to run; without it, the harness still works but those checks are advisory unless hard enforcement is configured. The package has an installation-script risk and a terminal surface: setup and CLI operations can install components and run commands with the agent’s available permissions. The mstar alias shares its npm bin namespace with an unrelated package, and co-installation can overwrite the shim.
Written from the project's own documentation and kept in sync with it. Where the two disagree, the source is authoritative — read the README on GitHub
Same category
Q00
Config-only bundle that mounts Ouroboros through the DSH MCP client, exposing 36 interview, Seed, execution, evaluation, and evolution workflow tools in DSH.
No one-line install — this plugin lives inside a larger repository and publishes no npm package.
loopx-project
LoopX, a provider-neutral, local-first state kernel and control plane for long-horizon agents: keeps Goal, Todo, gate, evidence, quota, recovery, and handoff state above DeepSeek Harness, while the plugin bootstraps the CLI and skills, admits bounded same-session continuation, and adds a loopback GoalBar for the exact bound loop.
No one-line install — this plugin lives inside a larger repository and publishes no npm package.
huangruiteng
LoopX, a provider-neutral, local-first state kernel and control plane for long-horizon agents: keeps Goal, Todo, gate, evidence, quota, recovery, and handoff state above DeepSeek Harness, while the plugin bootstraps the CLI and skills, admits bounded same-session continuation, and adds a loopback GoalBar for the exact bound loop.
No one-line install — this plugin lives inside a larger repository and publishes no npm package.
chuspeeism
Embeds the active installed Codex Taskboard runtime in the DeepSeek Harness sidebar, using its launcher runtime descriptor instead of a fixed port.
No one-line install — this plugin lives inside a larger repository and publishes no npm package.
tong-io
TongFlow film-crew studio for image, voice, music and video production: the agent writes per-asset TongFlow workflow files (.tongflow.json) that run through TongFlow plugins, with an embedded workflow canvas, a shot/character/take project layout and a manga-drama template; sessions starting with @tongflow open the Studio view.
No one-line install — this plugin lives inside a larger repository and publishes no npm package.
