跳到主要内容

dsh-agent-looper

已验证

@dancingteeth/dsh-agent-looper · v0.1.1 · MIT

DeepSeek Harness companion — teaches and scaffolds Agent Looper (GOAL.md + verify.sh + agent-loop CLI)

安装

dsh plugin add @dancingteeth/dsh-agent-looper

用 dsh --profile default --dump-config 确认 layer 已生效 —— 参见安装指南。

源码

发布到 npm 但没有公开仓库。安装前请检查包内容。

标签

作者

说明文档


tags:

  • agentic_ai
  • agents
  • documentation
  • loops
  • plugins

Agent Looper (DSH companion plugin)

Guidance for designing and wiring Agent Looper inside DeepSeek Harness (dsh web). This bundle does not install or replace the @dancingteeth/agent-looper CLI — it teaches the DSH agent how to use it.

What you get

Component Purpose
Skill design-loop Freeze measurable GOAL.md + verify.sh
Skill install-agent-looper npm install, init, scripts (not the grind)
Skill review-gate postQualityReview / reviewGate without thrash
Skill run-loop-in-dsh After freeze, start agent-loop as bash run_in_background: true
Command loop-scaffold Guided GOAL + verify scaffold (direct UI — not sent to the model)
Prompt + bash guard Always-on routing; block foreground agent-loop run; allow background jobs. runtime: dsh grinds need Full Access (nested headless writes ~/.dsh/profiles/headless/).

Shell verify (via agent-loop run) remains the finish line. DSH built-in /loop and /goal are not the exit wedge.

Install locally

Build JS first ("main" is dist/index.js — DSH will not load .ts). Use Node ≥ 22.15.

From this repository checkout (plugin package has no local tsc — use the repo TypeScript):

pnpm exec tsc -p plugins/dsh-agent-looper/tsconfig.json
dsh plugin --profile web add ./plugins/dsh-agent-looper
dsh web

Optional config on the plugin row in your profile patch:

- id: agent-looper
  name: '@dancingteeth/dsh-agent-looper'
  config:
    skillsDir: ./skills
    agentLoopBinary: agent-loop
    blockNestedRun: true

Runtime install (consumer repos)

pnpm add -D @dancingteeth/agent-looper
pnpm exec agent-loop-init

See skill install-agent-looper in DSH or plugins/agent-looper/ for the Cursor companion.

Docs