<div align="center">

dsh-automode ⚡

Claude Code–style auto mode for DeepSeek Harness — let your agent run hands-free, while a deterministic guardrail and a cost-aware reviewer keep the dangerous stuff from ever executing.

🌐 简体中文: README.zh.md · English: README.md

<img src="docs/auto-mode-icon.png" width="400" alt="dsh-automode in the permission picker" /></div>

---

dsh-automode sits between your agent and the harness. It intercepts every tool call before execution — hard deny rules and curated allowPaths decide deterministically (zero LLM cost), and whatever is left goes to a two-stage classifier. Safe actions run on their own; risky ones are blocked, reshaped, or routed to you.

✨ Key features

• 🛡️ Deterministic first line — regex deny bands hard-block exfiltration, secrets, and system paths before any LLM call; prefix-glob allow rules approve routine commands for free.
• ⚡ Zero-confirmation allowlist — config.allowPaths is full trust: file ops and bash writes inside it skip the classifier entirely, and escalated calls are auto-granted through the approval bridge (v0.10.0) — no prompt, no round-trip.
• 🧠 Cost-aware two-stage classifier — a one-token filter pre-screens; only flagged actions get the structured review, and identical actions reuse the verdict cache for 5 minutes.
• 🔁 Circuit breaker + human fallback — 3 consecutive (or 20 total) DENYs pause auto mode and route decisions to a human; one human decision resumes and resets.
• 📜 Full audit trail — every allow / deny / bridge decision is appended to ~/.dsh/auto-mode/decisions.jsonl.
• 🔌 Native preset — flip it on from the permission picker or /auto; it plays nicely alongside read-only / workspace-write / danger-full-access.

⚠️ Not a sandbox. The plugin runs inside the DSH process; a deliberately malicious plugin can do anything your user account can do. It reduces unsafe autonomous tool use — it is not an OS security boundary.

📚 Table of contents

• Install
• Commands
• How it works
• Rules
• Configuration
• Logging
• Architecture
• Compatibility & contributions
• License

Install

dsh plugin add @log.li/dsh-automode

From a local checkout:

dsh plugin add ./path/to/dsh-automode

Restart dsh web after installing. The permission picker (bottom-left of the chat box) will show Auto mode alongside read-only / workspace-write / danger-full-access.

Commands

/auto           # switch this session to auto mode
/auto-status    # show diagnostics: preset, approval policy, breaker state

How it works

Tool call arrives
  │
  ├─ [pre-execute gate]  (all tools, first defense)
  │    ① Read-only tools → allow (unless deny matched)
  │    ② Deny rules (regex) → hard reject
  │    ③ Allow rules (prefix glob) → approve
  │    ④ In-tree file ops → approve (allowInsideWorkingDirectory)
  │    ⑤ Escalation intent → classifier pre-screen
  │    ⑥ Everything else → pass through
  │
  └─ [approval waterfall]
       ① Soft deny rules (prose) → reject
       ② Soft allow rules (prose) → approve
       ③ Read-only allowlist → approve
       ④ Verdict cache hit → reuse (no second LLM call)
       ⑤ Classifier (two-stage: one-token filter → structured review)
       ⑥ Failure → fail-closed

🖱️ Interactive version: docs/auto-mode-flow.html — pan/zoom, relationship tracing, dark mode. Diagram source: docs/auto-mode-flow.workflow.json.

The pre-execute gate intercepts ALL tool calls (including those inside the workspace sandbox that would never trigger the approval waterfall). The approval waterfall only runs for calls that actually need sandbox escalation. The pre-execute gate only applies to auto-mode sessions; in other presets (read-only / workspace-write / danger-full-access) it is a no-op so it never contradicts the sandbox the user chose.

Rules

The rule system has two layers:

Hard boundary (deterministic, never goes to classifier)

• deny — regex patterns that hard-reject. First match wins. Evaluated before everything else. Use for exfiltration, secrets, sensitive targets, dangerous commands.
• allow — prefix-glob patterns that approve without any LLM call. Evaluated after deny. Use for routine commands you trust completely.

Classifier guidance (prose, fed to the LLM)

• rules.deny — soft-deny descriptions. The classifier reads these as standing rejections. Can be overridden by direct user intent or a matching allow rule.
• rules.allow — soft-allow exceptions. The classifier reads these as standing approvals that override matching soft-deny rules.
• rules.environment — context facts (trusted repos, infrastructure, cloud buckets). The classifier uses these to judge whether an action is within the user's environment.

All rules.* arrays support $defaults: using ["$defaults", "my custom rule"] keeps the built-in rules while adding yours. Omitting $defaults replaces the entire built-in list for that section.

Configuration

Configuration goes in your profile's cordis.patch.yml. Everything has defaults; a bare {} config is valid. The table below is the full reference; a minimal example (the allowPaths override) is under Trusting extra directories.

Key options

Option	Default	Description
deny	built-in list	Regex patterns that hard-reject. First match wins.
allow	built-in list	Prefix-glob patterns that approve without LLM.
readOnlyTools	read, glob, grep, list, search	Tools that default-allow (unless deny matched).
allowPaths	[]	Curated full-trust external directories: file ops and bash write-commands inside skip the classifier, and (since v0.10.0) escalated calls are auto-granted (approval-bridge). See Trusting extra directories.
allowInsideWorkingDirectory	true	Allow in-tree file ops without classifier.
classifier.provider / classifier.model	'' (follow session)	Override the classifier's LLM route. Resolution: classifier.{provider,model} → session's active model → agent's configured model.
classifier.reasoningLevel	off	Classifier reasoning effort (off disables reasoning). If a route rejects the effort, the call retries without it.
rules.deny	['$defaults']	Soft-deny prose for the classifier.
rules.allow	['$defaults']	Soft-allow prose for the classifier.
rules.environment	['$defaults']	Environment facts for the classifier.
failClosed	true	Reject on classifier failure vs. fall back to approval chain.
preExecuteGate	true	Enable the pre-execute gate (only applies to auto-mode sessions).
timeoutMs	45000	Per-call hard timeout for the classifier LLM calls.
classifyContextChars	6000	Char budget for the task-alignment context given to the classifier.
maxArgsChars	4000	Char budget of the command signature used for the verdict cache key.
breakerConsecutive	3	Consecutive classifier DENY to trip the breaker.
breakerTotal	20	Total classifier DENY to trip the breaker.

Trusting extra directories (allowPaths)

allowPaths is a curated full-trust list: file ops and bash write-commands whose target resolves inside one of these directories skip the safety classifier entirely (logged as pre-execute-allow / curated allowPath). The shipped default keeps only the universal /tmp/ — personal directories are configured per profile in your profile's cordis.patch.yml. Loader patches replace the targeted row's whole config, so the minimal override below sets only allowPaths (every other field falls back to the plugin's code defaults):

# ~/.dsh/profiles/<profile>/cordis.patch.yml
- id: auto-mode
  config:
    allowPaths:
      - /tmp/
      - /Users/<you>/Library/CloudStorage/OneDrive-<tenant>/Projects/<proj>/Proposal/

Only recognized write-commands are trusted (deletion such as rm/trash is never allowlisted), and paths are matched after symlink resolution — both a /Users/<you>/OneDrive - … symlink and the real Library/CloudStorage/… path work. A git add/commit/push inside an allowlisted directory resolves its repository root as the write target, so committing/pushing a repo that lives under an allowPath skips the classifier too (v0.11.1). The verdict-cache fix below still matters: even without an allowPath, once you explicitly authorize an action the classifier re-runs with your intent instead of replaying a cached denial.

Composite write-commands (v0.11.0). A temp→swap export dance like DIR=…; cp a b_tmp && (trash b; true) && mv b_tmp "$DIR/b" is now parsed segment-by-segment (with VAR=… assignment tracking and $VAR expansion), so its destinations still hit allowPaths. cd <dir> is a tracked benign navigator — it updates the effective working directory (so a later git add/commit/push resolves its repo root from it) without invalidating the fast path (cd - stays unpredictable and falls back). The fast path stays guarded: a composite containing a side-effect command (kill, pkill, rm, sh, bash, network/daemon management, …), a command substitution (`…` or $(…)), a file redirection (>file, >>file, 2>file), or any command outside the recognized write/benign set falls back to the classifier exactly as before. A descriptor-dup redirect (2>&1, >&2, >&-) is not a file write and does not invalidate the fast path — so git push … 2>&1 | tail is still allowlist-checked. Benign utilities inside a composite (trash, mkdir, echo, …) ride along with the allowlisted write — their side effects are not re-reviewed once every write target is allowlisted (deletion targets themselves are never allowlist-trusted, and rm still forces a classifier fallback). History-rewriting git commands (reset --hard, clean, rebase, merge) are deliberately not allowlist-trusted.

Zero-confirmation escalations (v0.10.0). An allowlisted path means full trust, so a call that asks to widen the sandbox (sandbox_permissions: danger-full-access) into an allowlisted path is now auto-allowed with no confirmation and no classifier — the pre-execute gate already proves every target sits inside an allowPath, and that verdict is carried to the approval answerer by the call's callId (the audit trail shows curated allowPath → approval-bridge → decision allowed-once). Deny patterns still run first (a deny-listed path such as ~/.ssh/ is hard-rejected even inside an allowPath), and the circuit breaker is never bypassed — while tripped, allowlisted calls still go to a human.

Not a file-sandbox exemption. allowPaths only skips this plugin's review — DSH's file sandbox (the session's file policy) is a separate layer and still applies. A write to an allowlisted directory outside the workspace is blocked by the sandbox unless the call asks for sandbox_permissions: danger-full-access; for an allowlisted path that escalation is auto-granted by the approval bridge (no review), so request the escalation on the first attempt.

Every auto-mode session also receives this knowledge as a system-prompt section (auto-mode:allowlist): the model knows where the per-profile allowPaths lives and how to edit it, so when an action is blocked it can propose the exact config change — and only applies it after you explicitly confirm.

Permission preset icon

The auto-mode preset ships a bolt glyph in the permission picker. Set your own logo by overriding icon on the preset (the permission row of cordis.patch.yml, not auto-mode's config) — an SVG path drawn inside the shield outline:

- id: permission
  config:
    presets:
      auto-mode:
        sandbox: workspace-write
        approval: ask
        name: Auto mode
        description: ...
        icon: '<your-svg-path-d>'   # default bolt: 'M9.15 3.4L5.85 8.55H7.95L7.05 12.6L10.45 7.25H8.25L9.15 3.4Z'

The icon is cosmetic — behavior is identical whether it renders. It shows only on DSH versions that read preset icons (stock DSH ignores it; the local dsh-permission-preset-icon.mjs patch enables it). Leave icon unset for the default bolt.

Two-stage classifier

The classifier uses two stages to minimize LLM cost:

1. One-token filter (~1 token): asks the LLM for a single digit (0 = safe, 1 = needs review). Uses a generous token budget and robust digit parsing so reasoning models aren't starved, and honors classifier.reasoningLevel as the reasoning effort (off = no reasoning). Most routine actions return 0 and skip stage 2.
2. Structured review: only runs when stage 1 flags the action. Returns a full verdict with reason.

This means most tool calls cost ~1 token of classifier overhead. Only borderline actions incur the full classifier cost.

The classifier is risk-based — it judges the action's real-world impact, not its surface form:

• Read-only and reversible operations are ALLOWED: GET/HEAD requests, inspection/listing/search/state queries, and local changes that can be safely undone (edits, temp files, builds, tests, git-tracked files).
• A sandbox-escalation request is NOT dangerous by itself — the classifier judges the action it enables. A reversible, low-blast-radius, user-aligned action may be allowed even when it needs escalation (e.g. editing a git-tracked skill or config file outside the working directory). Escalation for a genuinely dangerous action (exfiltration, persistence, weakening security, shared/production/external state) stays forbidden.
• <recent_user_intent> counts only direct human messages — with one deliberate exception: a user's answer to an ask_user_question tool call is a direct human authorization given through the tool, so it is folded into the intent window (and into the verdict-cache key, so a tool-based grant invalidates a stale DENY just like a typed message). Ordinary tool results, plugin/system injections, and model messages stay excluded, so the user's real instructions are never crowded out and only a human can grant permission.

Circuit breaker

When the classifier denies 3 actions in a row or 20 total in a session, the breaker trips and auto mode pauses. The approval waterfall delegates to the human answerer. Any human decision (allowing OR rejecting the action) resumes auto mode and resets all counters — a human in the loop breaks the silent-denial loop the breaker exists to catch. If the human cancels the request or no answerer is available, the breaker stays tripped.

At the moment the breaker trips, the plugin injects a hint telling the model to request danger-full-access sandbox escalation directly on its next attempt (which surfaces the human approval window immediately), instead of the "try at current level → hit a denied error → then escalate" round-trip.

Classifier failures (timeout, parse error, empty response) are NOT counted toward the breaker. Cached denies count, however — a repeat of an identical previously-denied action (verdict cache hit) increments the consecutive and total counters, so repeated escalation attempts actually trip the breaker and reach the human instead of spinning forever.

Denial guidance & diagnostics

The classifier is two-state (allow / reject) — there is no "ask" tier. An uncertain action is rejected (fail-closed): a rejected action can be retried in a safer form or escalated to the user, but a wrongly-allowed action cannot be undone.

Routine categories (installs, builds, tests, file edits, git add/commit/status) are a tendency, not a free pass — the classifier must judge the specific command and arguments, never the category label alone (e.g. pipe-to-shell downloads, unknown-package installs with arbitrary postinstall scripts, secret writes, irreversible deletes, pushes to unknown remotes).

When an action is denied, the denial echoes the reviewer's reason AND the model's own stated justification (from the tool call), so the model can see exactly what was rejected and reshape it. It is told to try a safer alternative; if no safer alternative exists, it is instructed to stop retrying and ask the user for explicit permission — a denied action will keep failing, and only explicit user approval lets a later attempt pass (the classifier weighs the user's recent explicit intent via <recent_user_intent>).

Every classifier stream failure (thrown error or an error finish chunk) is logged to the DSH log with the resolved route, effort, error code/message, and raw output, and written to decisions.jsonl as a classifier-fail event — so a recurring classifier returned no verdict is diagnosable from the audit log itself. If a route rejects the configured reasoningEffort (e.g. low on a route that only supports off), the call is retried without an effort before failing.

Verdict cache

Classifier verdicts are cached per session by tool + command + user-intent signature. The user's recent direct instructions — typed messages and ask_user_question answers — are hashed into the key, so a new explicit authorization (a fresh human message or a tool-based grant) invalidates a previously cached verdict and the classifier re-runs with the new intent — a user grant is never swallowed by a cached DENY. Within the same intent window a repeated action still reuses the cached verdict without a second LLM call. Cache entries expire after 5 minutes.

Logging

All decisions are logged to ~/.dsh/auto-mode/decisions.jsonl (JSONL format, append-only, survives restarts). Each entry includes:

• at — ISO timestamp
• event — decision / pre-execute-deny / pre-execute-allow / pre-execute-fileop / pre-execute-bashop / pre-execute-fail-open / classifier-fail / breaker / resume / boot
• outcome — allowed-once / rejected / cancelled
• tool — tool name
• tier — deny / allow / classify:monitor / classify:cache / classify:fail / ...
• detail — human-readable reason
• sessionId — session identifier

Use the review script to analyze the log and identify rule optimization opportunities:

node scripts/auto-mode-review.mjs

System prompt shadowing

When auto mode is active, the plugin shadows the approval-policy system prompt so the model sees "auto" instead of "ask". This tells the model that tool rejections come from the automated reviewer, not from a human. The model adjusts its retry strategy accordingly (try a smaller/safer action instead of asking the user).

Architecture

src/
  index.ts         Main entry: preset management, approval answerer, breaker reset, commands, system-prompt shadowing
  config.ts        Config schema + $defaults mechanism + built-in rule lists
  bands.ts         Deterministic band engine (deny regex + allow glob)
  pre-execute.ts   Pre-execute gate (first defense; real-path trust, classifier pre-screen, breaker-trip hint)
  classifier.ts    Two-stage classifier (+ robust parser, reasoning effort, diagnostics)
  rules.ts         Prose rule matching for the classifier
  prompt.ts        Classifier prompt construction (<recent_user_intent> + intent weighting)
  cache.ts         Verdict cache (shared across enforcement points)
  breaker.ts       Circuit breaker (3 consecutive / 20 total)
  log.ts           Shared appendDecision JSONL logger

Compatibility & contributions

Supported dsh versions: 0.1.0-rc.6 – 0.1.x (peer range >=0.1.0-rc.6 <0.2.0). Permission facts (preset / sandbox / approval) are read through two auto-detected paths:

dsh version	permission read path
0.1.0-rc.6 – 0.1.4.x	session.events event log (effectivePermissionPreset & co.)
≥ 0.1.5-rc.1	durable permissions session projection (ctx.sessionProjections.stateOf)

dsh 0.1.5-rc.1 removed the session.events accessor — without the projection path every auto-mode turn died with Cannot read properties of undefined (reading 'length') while the system prompt was assembled (fixed in v0.11.2). dsh ≥ 0.2.0 is not yet verified — bump the peer range only after testing against a new core.

• Verified on macOS only. Tested against the macOS filesystem, the DeepSeek Harness (DSH) runtime, and the DSH version in use at development time. Path semantics — including the macOS /tmp → /private/tmp symlink (handled by realpath-nearest-ancestor resolution) and workspace-path trust — have not been verified on Linux or Windows, and deny-pattern/path matching may differ there.
• Found a bug, or an issue on another platform? Bug reports and pull requests are welcome — open an issue or PR at github.com/log-li/dsh-automode.

License

MIT