dsh-regression
Run the following command in DeepSeek Harness:
dsh plugin install chenghaoYang/dsh-regression
Paste the following prompt into your AI chat to install this plugin:
Install it in DeepSeek Harness by running dsh plugin install chenghaoYang/dsh-regression; the source is at https://github.com/chenghaoYang/dsh-regression
About this plugin
Coding-agent output quality has long relied on manual review or letting another model act as judge, neither stable nor transparent. dsh-regression turns an explicit correction into a repeatable, deterministic regression case so that whether the workspace actually violated the contract stops depending on human intuition or probabilistic guessing.
The plugin runs the same task inside an isolated Git worktree and checks the final workspace with four deterministic verifiers: command exit code, diff-path boundary with file-count caps, json-schema structural validation, and api-snapshot baseline comparison. It never invokes a second model as judge. It also supports comparative reporting across two DSH profiles and a declarative Cause analysis that uses environment overlays plus delta debugging to isolate the 1-minimal set of components responsible for a failure.
If your team already drives coding agents through DeepSeek Harness and wants CI-grade, reproducible verification of post-edit workspace compliance including cross-profile comparison, failure attribution, and a clean DSH command surface, dsh-regression is the workflow layer built for that. It needs no extra API key, and the bundled five-case verifier smoke pack exercises the full pipeline locally out of the box.
Use Cases
- After a coding agent edits code, deterministic verifiers automatically check whether the workspace violates path boundaries, file-count caps, or schema contracts
- Run the same regression case under two DSH profile configurations and generate a comparative report to pinpoint regressions
- Apply declarative environment overlays with delta debugging to isolate the 1-minimal set of components causing a failure
Best For
- Teams driving coding agents via DeepSeek Harness that need CI-grade regression verification
- Developers who want to solidify a one-off manual correction into a repeatable executable case
- Platform engineers comparing multiple agent configurations and performing failure attribution
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.