Skip to main content

ouroboros/integrations/dsh-plugin

5.6kStars559Forks59Issues14Watchers

Launch the Ouroboros Python engine via uvx to provide DSH with specification-driven AI workflows as MCP tools (Interview→Seed→Execute→Evaluate→Evolution loop), requiring no installation or additional code.

Evidence5/5methodologySourceInstallMaintenanceDSH versionSecurity scan
Machine-auditedInstall commandRepo verifieddsh-plugin topicLicenseREADMEAI wiki
Language
Python
License
MIT
Branch
main
agent-osagentic-aiai-agentai-coding-agentclaude-codeclicodexcoding-agent

Install

cmdweb profile
$ dsh plugin --profile web add github:Q00/ouroboros#path:integrations/dsh-plugin

Run the command above in your terminal to install this plugin via the dsh CLI. You can switch Profile in the top-right corner. New to dsh? Read the beginner tutorial

Install via your agent

Install the DeepSeek Harness plugin Q00/ouroboros/integrations/dsh-plugin for me: review the repository at https://github.com/Q00/ouroboros first, then run the install command and verify the plugin loads successfully.

Paste this instruction to the DSH Web GUI assistant — it will install and verify for you.

One-Line Pitch

This bundle integrates Ouroboros—a "clarify requirements before coding" Python AI workflow engine—into DeepSeek Harness as an MCP tool, eliminating window switching when you type ooo interview, ooo auto, or ooo qa in the dsh dialog.

Core Capabilities

  • Directly invoke 30+ Ouroboros tools (interview, Seed, execute, evaluate, evolve, QA, undo, rollback, etc.) in dsh conversations—the model auto-selects based on tool descriptions
  • Run the complete interview → Seed → execute → evaluate loop (auto tool runs through in one pass synchronously, 30-minute timeout)
  • Use Socratic interviewing within existing dsh projects to reduce requirement ambiguity, only allowing Seed spec generation after reaching threshold
  • Refine Seed through evolution loops (evolution, rollback, generation tracking) until the evaluator approves
  • Exception branch tools: cancel execution, cancel background tasks, query task status, query projection events, query dashboard
  • Soft failure startup: if uv isn't installed or Ouroboros isn't configured, dsh still starts normally—just Ouroboros tools won't display, other plugins unaffected

Technical Implementation

  • Language: The entire bundle is pure config (YAML + JSON), no custom plugin code; Ouroboros itself is Python
  • Key Dependencies: @deepseek-ai/dsh-mcp-client (MCP stdio client), uvx (on-demand Python runtime), ouroboros-ai[mcp] (MCP 2.0 protocol server package)
  • Architecture Pattern: Injects one line mcp-ouroboros via Cordis patch (dsh.bundle.patch), dsh's MCP client spawns ouroboros mcp serve via stdio protocol, sub-process exposes 30+ tools to the upper agent
  • Entry Files: integrations/dsh-plugin/cordis.patch.yml (logic entry), integrations/dsh-plugin/package.json (declares dsh.bundle.patch path)

Use Cases

For those who want DSH to stop "jumping straight into code": when using dsh for real requirements, first let Ouroboros break vague ideas into acceptance-ready Seeds, then hand to the executor, and finally let the evaluator score and iterate. Also suitable for developers who need to run research, design, and review workflows in dsh without leaving the dsh window.

Prerequisites & Compatibility

DependencyMin VersionNotes
Python>= 3.12Auto-fetched by uvx in isolated environment, no manual pip install
uv (uvx)Latest stableMust be in PATH; on first launch it downloads Ouroboros and its dependencies
dshNo version declaredBundle itself has no constraints; parent repo loads via dsh-mcp-client
Node.js>= 22Only needed when using OUROBOROS_LLM_BACKEND=dsh and building dsh from source
Agent runtimeCLI runtime executableAny of claude-cli, codex, opencode; tells Ouroboros who executes ooo auto
OSCross-platformNo native modules, not locked to Linux/macOS/Windows

Installation

dsh plugin --profile web add github:Q00/ouroboros/integrations/dsh-plugin

Configuration

This bundle has no DSHP-style "config page"—all optional parameters are injected via environment variables, dsh sub-process only allows these to pass through:

ConfigTypeNotesDefault
OUROBOROS_AGENT_RUNTIMEenv varSpecifies which CLI runs ooo auto execution steps (claude-cli, codex, opencode, etc.)Not set, inherits from ouroboros setup default
OUROBOROS_LLM_BACKENDenv varWhich LLM to use for interview/Seed/QA; set to dsh to route through DeepSeek Harness (requires below two)Not set, inherits from ouroboros setup default
OUROBOROS_DSH_CONFIG_PATHenv varRequired when OUROBOROS_LLM_BACKEND=dsh, absolute path to dsh's Cordis composite configNot set
OUROBOROS_DSH_CLI_PATHenv varPath to dsh-acp-demo binary when not in PATHUses PATH
ANTHROPIC_API_KEYenv varPass through to Ouroboros default LLM backendEmpty if not set
DEEPSEEK_API_KEYenv varLoopback credential for dsh backendEmpty if not set
toolCallTimeoutMspatch paramouroboros_auto sync input, 30 minutes (1800000 ms)Already written in patch

Other non-credential variables (PATH, HOME, OUROBOROS_* selectors) are automatically passed through by dsh—no need to declare them here.

FAQ

Q: Must I install Python?

A: System-level Python not needed. The bundle uses uvx to download and run Python 3.12+ with Ouroboros in an isolated environment—just need uv on the machine (one-line install from https://astral.sh/uv).

Q: Can't see mcp__ouroboros__* tools after installation?

A: First confirm uv --version runs; then confirm whether you set OUROBOROS_AGENT_RUNTIME or ran ouroboros setup. Remember startup failure is soft (failOnStartupError: false)—dsh won't crash from this plugin, just missing from the tool list.

Q: What's missing compared to installing Ouroboros separately?

A: No SKU difference—the bundle is essentially one line of cordis.patch.yml, the real Ouroboros package still pulled from PyPI via uvx. In other words, you can chat in dsh or run ooo standalone in terminal, both share the same config.

Q: How do I pass in OpenAI/Google keys?

A: Copy the entire mcp-ouroboros config section in your profile's cordis.patch.yml, add OPENAI_API_KEY (or GOOGLE_API_KEY, etc.) in env; note dsh's later layer overwrites the entire section rather than deep merging, so don't patch just one line—rewrite with the full config block.

Q: Network down/uv acting up, tool list still there but calls keep failing?

A: Expected behavior: this bundle has no auto-reconnect logic (reconnect loop only exists in dsh main repo main branch), you need to fix the root cause then manually restart dsh or reload plugin.

Learning Curve

Advanced — no coding required, but need to understand uv, environment variables, and Ouroboros's agent runtime selection layer to enable smoothly; average users need to read the README before use.

Known Issues & Limitations

  • OUROBOROS_LLM_BACKEND=dsh is not a single variable switch: must build dsh from source (pnpm install && pnpm run build, Node.js >= 22), released @deepseek-ai/dsh-acp-demo still conflicts on dsh-tool-bash peer dependency chain; also must provide absolute path to Cordis composite file (OUROBOROS_DSH_CONFIG_PATH), relative paths rejected
  • MCP sub-process can't host in-process Claude SDK, so OUROBOROS_AGENT_RUNTIME must be an executable CLI (can't fill SDK names like claude), unless you want ouroboros auto step to error on invocation
  • Sub-process environment has credential stripping: only ANTHROPIC_API_KEY and DEEPSEEK_API_KEY are whitelisted, other *_KEY variables won't auto-pass through to Ouroboros
  • Reconnect is not default behavior: released dsh 0.0.1-rc.1 has no reconnect loop, only latest main branch has version with exponential backoff and retry cap; after install if it crashes, need to manually reload plugin or restart dsh
  • Bundle doesn't publish independent version number or release, follows parent Ouroboros repository's main branch, meaning dsh plugin add always gets upstream main head

Read the usage guide →

Install steps, key points, FAQ and compatibility for this plugin — auto-derived from indexed fields.

Listing badge

Listed on deepseek-plugin.org
[![Listed on deepseek-plugin.org](https://img.shields.io/badge/listed_on-deepseek--plugin.org-007EC6)](https://deepseek-plugin.org/plugins/Q00/ouroboros/integrations/dsh-plugin)

Paste this markdown into your GitHub README to link back to this listing. The badge only states the listing — not a security endorsement.

← Back to plugin directory