dsh-trajectory-teacher
Run the following command in DeepSeek Harness:
dsh plugin install XiaoMoDern/dsh-trajectory-teacher
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install XiaoMoDern/dsh-trajectory-teacher in your DeepSeek Harness terminal, then restart DSH to auto-mount the plugin; source code is available at https://github.com/XiaoMoDern/dsh-trajectory-teacher .
About this plugin
Every time an Agent finishes a task run, the raw trajectory is a wall of hundreds of token-level chunks interleaved with tool calls, approval events, and error logs. Trying to figure out which step was smart and which was a stumble by scrolling through that stream is basically impossible.
dsh-trajectory-teacher solves exactly that: type /trajectory-teacher in the chat, and the plugin extracts the key events from the current session, filters out the noise, feeds a concise summary into an LLM, and returns a structured Markdown retro covering decision points, pitfalls, and reusable patterns. Sensitive fields such as API keys are automatically redacted before output. The whole flow is manually triggered, length-capped, and ships with deep reasoning turned off by default, keeping token costs predictable.
Built for developers who drive multi-step Agent tasks in DeepSeek Harness on a daily basis and want one command to turn a black-box run into a readable post-mortem.
Use Cases
- Generate a structured retro after each multi-step task run with one command
- Pinpoint which step in an Agent run went wrong and document the pitfall
- Accumulate reusable decision patterns into team-readable docs
Best For
- Developers who drive multi-step Agent tasks in DSH on a daily basis
- Tech leads who need to audit Agent decision chains after the fact
- AI engineers who want to turn black-box runs into readable retros
Related Plugins
A method pack that makes AI coding agents plan against your real baseline, prove completion with fresh evidence, and reduce reworks and unsafe changes.
Turns the DeepSeek Harness session into a captain that builds a durable sub-agent team, splits goals into dependency-aware tasks, and coordinates work via direct messages and a live Web UI.
Gives coding agents design judgment, letting Claude Code, Cursor, and 70+ agents generate and iterate high-quality UI, presentations, and graphics right from the terminal.
Run the Pi ecosystem's plugins on DeepSeek Harness, unmodified, via a compatibility layer that implements Pi's public extension ABI on DSH's native services.