Bear0117/dsh-plan-checkup

Reads a plan step by step before you approve it and marks what it finds on the plan review card: irreversible actions, effects outside the workspace, unrequested work, vagueness and missing verification.

The flags are split between mechanical rules and judgment: DROP TABLE, rm -rf and git push --force are caught by command rules, as are git push, npm publish and kubectl apply for the outside-the-workspace flag, while unrequested work, vagueness and missing verification come from the judgment engine. The badge in the review card toolbar shows progress while the engine walks the steps, then the count of flags per colour or 'No issues found'; the details panel lists each flagged step with probabilities, the commands a rule matched, and whole-plan notes. It only advises — it never blocks, rewrites or approves a plan — and nothing leaves the machine until you point it at an OpenAI-compatible or Jev-compatible endpoint, with loopback servers exempt from allowEgress. Each flag has thumbs up/down buttons kept in a local file for later threshold recalibration.

Agent Capabilities ★ 1
View on GitHub ↗

Install

dsh plugin --profile web add dsh-plan-checkup

The npm line from the README; dsh-plan-checkup@0.1.1 resolves its registry repository to this repo. Run the same line through npx @deepseek-ai/dsh if you run dsh via npx, then restart dsh web.

Compatibility

dsh 0.2.x, tested by the author on 0.2.0-rc.2 with the web profile on Windows 11 and Node.js 24.21. Until a judgment engine is configured only the command rules run and the badge says 'Rules only'. pnpm only picks a newly published version once it is a day old, so an update in the first day needs the version named explicitly.

Details

Alternatives

syyr1987/dsh-linghun · gy-bai/dsh-anchored-subagent · memorax-ai/dsh-patchouli

More plugins in Agent Capabilities

Browse more in Agent Capabilities

Guides for Agent Capabilities plugins