dsh-tool-strict-check
Run the following command in DeepSeek Harness:
dsh plugin install catsenior507/dsh-tool-strict-check
Paste the following prompt into your AI chat to install this plugin:
Run `dsh plugin install catsenior507/dsh-tool-strict-check` inside your DeepSeek Harness profile to install this plugin; the full source is at https://github.com/catsenior507/dsh-tool-strict-check .
About this plugin
An agent that re-reads its own diff gets a longer version of the same mistake. The failures that actually burn turns—a file that no longer parses, a shell construct the harness shell cannot run, a sorry hiding in a Lean proof, a type error three files away—are mechanically decidable. They do not need judgement; they need a compiler.
strict_check hands the work to each language's own checker. The batch tier runs py_compile, node --check, tsc --noEmit, and the PowerShell parser. The commands tier applies static rules over shell text without ever executing it. Status probes what is available on the machine and reports honestly. The most rigorous tier is Lean: hasSorry promotes unfinished goals to failure, -DmaxErrors=0 lifts the default hundred-error cap, -DautoImplicit=false closes the footgun where a misspelled hypothesis silently becomes a fresh universally quantified variable, and the toolchain version is anchored at 4.33.1 because eight soundness bugs before that release each allowed the kernel to accept a proof of False. When a checker is missing, the result says unavailable and prints the path it looked for; it never disguises an unverified file as a pass.
Built for developers working in DeepSeek Harness who write multi-language code or Lean proofs and want every agent output adjudicated by a real compiler rather than by the model re-reading its own words.
Use Cases
- Auto-validate agent-generated code with py_compile, tsc, and other real compilers
- Statically scan shell commands to block constructs the harness shell cannot execute
- Verify Lean proofs contain no sorry and the toolchain meets the soundness floor
Best For
- Developers writing multi-language code in DeepSeek Harness
- Researchers working with Lean formal proofs
- Teams that require agent output adjudicated by real compilers rather than self-review
Related Plugins
A network security red-team benchmark plugin for DeepSeek-V4.1/Flash featuring dual-layer kernel injection and zero-tool architecture for controlled model safety assessment.
A security red-team prompt plugin for DeepSeek-V4 that uses code-formal mapping and zero-trigger system prompts to enable unconditional output, intended for authorized testing and research only.
Adds an Auto approval mode on the official workspace-write sandbox, classifying semantic risks, asking when ambiguous, and denying destructive operations.
Undo/rollback system for DSH: config & plugin snapshots, one-click undo/redo/restore, message-level revert, secret masking, safe mode, and offline WebUI/GUI/CLI rescue tools for when DSH won't start.