Violet-Evergarden-Xiya/deepseek-AutoReview

README 摘要

deepseek-autoreview A「替我审核」(review-on-my-behalf) permission preset for DeepSeek Harness: a fourth permission mode beside read-only / workspace-write / danger-full-access . Escalation asks run through a three-tier funnel — a conservative whitelist (zero-token fast approval), a danger blocklist (zero-token interception), and an independent reviewer model for the gray zone. Safe operations auto-approve, unsafe/uncertain ones fall through to the human approval prompt, and obviously malicious asks can be denied outright. Fail-closed by construction, with a user-switchable reviewer model in General settings. The four modes Preset Sandbox Approval Inside workspace Escalations read-only read-only ask writes denied human approval workspace-write workspace-write ask ✅ silent human approval review (替我审核) workspace-write ask ✅ silent whitelist → blocklist → reviewer model; safe auto-approves, else human/deny danger-full-access danger-full-access never ✅ silent ✅ silent Install Requires a dsh profile ( /.dsh/profiles/ , new dsh.profile.bundles + cordis.patch.yml structure). Then register the bundle in the profile's package.json ( dsh plugin add only installs the dependency): Restart dsh (no hot…

分类