How sallespro/dsh-dynamic-agent works
This repository provides two command-line workflows over DeepSeek Harness: a one-turn web-research agent and a structured two-stage extract-then-write chain. Both scripts author a fetch plugin and patch at runtime, launch the harness through its SDK client, and use an OpenAI route. The chain additionally feeds validated extraction results into a writing stage. The harness internals are external to this repository; links to their behavior are based on README documentation and the shown launch/configuration code.
Main entry points: dsh-chain-agents.mjs Chain runner · dsh-dynamic-agent.mjs Research runner
Components 18
CLI workflows
- Research runner
dsh-dynamic-agent.mjs - Chain runner
dsh-chain-agents.mjs
Runtime setup
- Fetch plugin
dsh-dynamic-agent.mjs - Runtime patch
dsh-dynamic-agent.mjs - Chain fetch plugin
dsh-chain-agents.mjs - Chain patch
dsh-chain-agents.mjs
Research runtime
- Harness SDK client
dsh-dynamic-agent.mjs - dsh runtime
- Research subagent
- fetch_url tool
dsh-dynamic-agent.mjs
Structured chain
- Harness SDK client
dsh-chain-agents.mjs - Workflow pipeline
dsh-chain-agents.mjs - Extract stage
dsh-chain-agents.mjs - Write stage
dsh-chain-agents.mjs - dsh runtime
Other components
- CLI user
- HTTP(S) pages
- OpenAI model
Connections 27
- CLI user → Research runner: runs prompt
README.md - Research runner → Fetch plugin: writes plugin
dsh-dynamic-agent.mjs - Research runner → Runtime patch: writes overlay
dsh-dynamic-agent.mjs - Research runner → Harness SDK client: launches and runs
dsh-dynamic-agent.mjs - Harness SDK client → dsh runtime: starts subprocess
dsh-dynamic-agent.mjs - Runtime patch → dsh runtime: configures runtime
README.md - Fetch plugin → dsh runtime: mounts plugin
README.md - dsh runtime → Research subagent: registers subagent
README.md - Research subagent → fetch_url tool: uses allowed tool
README.md - fetch_url tool → HTTP(S) pages: fetches URL
dsh-dynamic-agent.mjs - dsh runtime → OpenAI model: routes requests
README.md - OpenAI model → Research subagent: delegates research
README.md - CLI user → Chain runner: runs URL list
README.md - Chain runner → Chain fetch plugin: writes plugin
dsh-chain-agents.mjs - Chain runner → Chain patch: writes overlay
dsh-chain-agents.mjs - Chain runner → Harness SDK client: launches harness
dsh-chain-agents.mjs - Harness SDK client → dsh runtime: starts subprocess
dsh-chain-agents.mjs - Chain patch → dsh runtime: configures runtime
README.md - Chain fetch plugin → dsh runtime: mounts plugin
README.md - Chain runner → Workflow pipeline: calls pipeline
dsh-chain-agents.mjs - Workflow pipeline → Extract stage: runs stage
dsh-chain-agents.mjs - Extract stage → dsh runtime: runs agent
dsh-chain-agents.mjs - Extract stage → fetch_url tool: retrieves page
README.md - Extract stage → Write stage: passes structured facts
dsh-chain-agents.mjs - Workflow pipeline → Write stage: runs stage
dsh-chain-agents.mjs - Write stage → dsh runtime: runs agent
dsh-chain-agents.mjs - dsh runtime → OpenAI model: routes requests
README.md