v0.2.0

dsh-auto-reviewer v0.2.0

Codex-style "approve for me" permission mode for DeepSeek Harness.

Highlights

• dsh 0.2 support — peer ranges now cover 0.2.0-rc.x, so the bundle is no longer skipped by the 0.2 compatibility gate. Notices use a producer-owned source kind (dsh-auto-reviewer) as the 0.2 session format requires, and the permission-icon patch follows the glyph table that moved to @deepseek-ai/dsh-client-ui-permission-presets.
• Notice row visible again on dsh 0.2 — new client half (client/client.js). The 0.2 Chat drops plain injected-context rows from the transcript, so the plugin now owns a conversation node kind and renders the decision itself: the collapsed row reads "Auto-review approved / Auto-review denied" plus the one-line summary (approved(low) · bash · workspace-write · npm test), and the expanded body carries the full detail. On earlier dsh generations the built-in context row still shows the message and the client half skips its registration, so no duplicate row appears.
• Structured notices, the multi-level safety policy, the optional LLM review, and the permission-icon auto-patch are unchanged from v0.1.0.

Install

dsh plugin --profile web add github:AntaresCorn/dsh-auto-reviewer

Then fully restart DeepSeek Harness (systemctl --user restart dsh-web or equivalent), reload the browser page, and select auto-review in the permission picker. Notices already recorded in existing sessions render as rows after the reload.

Verification

• npm run build && npm test — 16/16 node:test cases: host notice formatting and every decision branch, plus the client-half bundle (node kind, decision parsing, and the 0.2 Chat visibility contract).
• Cold start on dsh 0.2.0-rc.2: no skipping profile bundle warning, the permission catalog lists auto-review, and the client bundle is served from the boot graph.
• Live run: both the auto-approve and the auto-deny path wrote a notice into the session, which the plugin node now carries into the transcript.

v0.1.0

dsh-auto-reviewer v0.1.0

Codex-style "approve for me" permission mode for DeepSeek Harness.

Highlights

• auto-review permission preset — extends the official permission table with workspace-write + ask; selectable in the UI permission picker.
• Automatic approval reviewer — prepended to the approval waterfall: safe sandbox escalations are auto-approved (allowed-once), risky/ambiguous ones ask the user, critical destructive operations are rejected unless the user explicitly confirmed.
• Optional LLM review — ambiguous requests can be reviewed by the session model; unavailable/timeout falls back to human confirmation (fail-safe).
• Permission icon auto-patch — adds the shield + sparkle auto-review glyph to the installed conversation bundle; now compatible with both dsh >= 0.1.2-rc.1 (new Map([...]) glyph entries) and legacy object-literal bundles.
• Structured approval notices — multi-line detail (outcome / risk / authorization, tool, requested sandbox mode, decision basis, command) injected with source.form = 'notice', so the collapsed transcript row shows a one-line summary such as approved(low) · bash · workspace-write · npm test.
• permissionPresets.current(session) API compatibility for dsh 0.1.2-rc.1.

Install

# via GitHub
dsh plugin --profile web add github:AntaresCorn/dsh-auto-reviewer

# or from the attached tarball
dsh plugin --profile web add ./dsh-external-dsh-auto-reviewer-0.1.0.tgz

Then fully restart DeepSeek Harness (systemctl --user restart dsh-web or equivalent) and select auto-review in the permission picker.

Verification

• npm run typecheck && npm run build
• npm test — 7/7 node:test cases covering the notice formatters and every decision branch (fast path, critical-rule deny, blocklist deny, ambiguous fallback).

Compatibility

Peer ranges @deepseek-ai/dsh-* >=0.0.1-rc.1 <0.1.0 || >=0.1.0-rc.1 <0.2.0-0; verified against dsh 0.1.2-rc.1.

License

BSD-3-Clause.