📖 Ecosystem knowledge base — measured data, not marketing: plugin development guide · plugin-selection data · maintenance criteria.

⭐ 如果它帮到了你

这个插件是 DSH 插件家族的一员（40+ 个，全部 Apache-2.0）。如果你在用，给个 star —— 它不会解锁任何功能，但会让下一个人在搜索里更容易找到它。

English: part of a 40+ plugin family for DeepSeek Harness. If it is useful, a star helps the next person find it — nothing is gated behind it.

Compatibility

What you get

dsh-permission-rules puts an ordered allow / deny / ask rule list in front of every tool call on the tools/pre-execute waterfall — deterministic, instant, auditable, and written by you in plain YAML:

• deny blocks the call; the rule's reason becomes the model-visible error.
• ask rides the official approval seam (mount dsh-auto-review for a second-model answerer, or a human answers; with neither, the harness fails closed).
• allow (and no-match) strictly delegates via next() — downstream listeners are never short-circuited.

Every hit and every passthrough is audit-logged as a permissionRules/decision session event (log-only — nothing extra is injected into the model context).

• Rich matching — tool-name globs (including mcp__*), agent-identity selectors (main / subagent / preset:*), argument key/value globs or regexes (with !pattern negation and an absent key dimension), workspace-relative path globs at any nesting depth, when host conditions (env vars, platform), and shell command decomposition (argv: command word, argument tokens, pipeline signature) for token-precise command matching.
• Built-in high-risk baseline — a shipped deny/ask ruleset (destructive commands, privilege escalation, download-and-execute, sensitive paths) enabled by default and appended after user rules so a nearer user rule can override it; toggle with builtin.enabled.
• Hierarchical rule files — optional searchUp merges every .dsh/rules.yaml from the session cwd to the filesystem root, nearest first.
• Dry-run rollout — enforce: false audits what the policy would do while passing every call through.
• Hot reload — Chokidar watch with debounce; a broken edit keeps the previous rules, never crashes. On a WSL host, or for a rule file under /mnt/<drive>, the watch switches to polling because native change events are unreliable there.
• Fail loud — invalid YAML, unknown actions/fields, bad globs/regexes, backtracking-prone patterns, or more than maxRules rules fail the load.

Rule syntax

# <project>/.dsh/rules.yaml
rules:
  - match: { tools: [bash, pwsh], params: { command: "git push*" }, paths: ["**/secrets/**"] }
    action: deny
    reason: "No pushes from protected paths"

  - match: { tools: [edit, write] }
    action: ask
    reason: "File writes need confirmation"

• Match dimensions — tools (globs, incl. mcp__*), agents (main / subagent / preset:<name>; unknown identity never matches — fail closed), params (key/value globs or regexes, !pattern negation, absent key dimension), paths (workspace-relative globs extracted at any nesting depth), when (env var globs/regexes + a closed platform list), and network (domains / ips / ports / schemes — globs, wildcards, CIDRs, port ranges).
• Actions — allow / deny / ask, evaluated in file order, first match wins.
• Rule metadata — enabled: false (visible but inert), description, tags; unknown fields fail the load.
• Schema — a JSON Schema ships at docs/rules-format.schema.json (editor completion via # yaml-language-server: $schema=...); the full vocabulary and a 5-rule security baseline live in docs/rules-format.en.md.

Network policy

A Codex-style process-level network policy: shell subprocess traffic flows through a built-in local HTTP/CONNECT proxy, and every connection is decided by ordered network rules or by three modes mapped onto the official sandbox presets:

• deny-all — the read-only sandbox preset: block all outbound.

• whitelist — the workspace-write preset: allow listed targets, unlisted: ask (or deny) for the rest.

• allow-all — the danger-full-access preset: allow everything.

• auto (default) — follows the sandbox preset; on hosts without the sandbox-policy service it resolves to autoFallback (allow-all).

• Matching — match.network with domains / ips / ports / schemes (globs, wildcards, CIDRs, port ranges; numeric YAML ports are accepted). URL-candidate extraction on the tools/pre-execute hot path fires on web-tool arguments and URLs embedded in bash/pwsh command text; loopback targets can short-circuit rules per loopback policy. IPv4-mapped IPv6 literals are normalized to their IPv4 form before matching, and the proxy connects on the addresses the decision was made on — never a second DNS resolution — a connection whose adjudication resolved no address fails closed (502) instead of dialing the name.

• Upstream chaining — network.upstreamProxy (default off) sends the connections this proxy allows through an upstream proxy: CONNECT asks the upstream for a tunnel (CONNECT host:port) and plain-HTTP requests are forwarded to it in absolute form. A blocked target never reaches the upstream — it still gets this plugin's structured 403. Two cases never chain even when an upstream is configured: a loopback target (a proxy outside this host cannot route its loopback) and any decision produced by an ips-scoped rule (chaining hands the hostname to the upstream, so "the connection lands on an address the rules saw" — issue #21 — would stop holding exactly where the rules cared about the address; those decisions keep dialing the adjudicated address directly), plus a target whose scheme has no usable upstream. The upstream's own hostname is operator configuration, not agent input, and is not adjudicated by these rules. A URL carrying credentials is never emitted raw: warnings, /rules network and the settings snapshot mask the password (http://user:***@host:port). An unreachable upstream, a timeout (10 s), or a non-2xx answer yields 502 — there is deliberately no silent fallback to a direct dial, so a misconfiguration stays visible.

• Audit — denied connections append permissionRules/network to the owning session (same adaptive ignorable gate), with block counters and recent interceptions in /rules network and the settings page.

• Diagnosis — blocked connections carry a [network: …] message naming the blocked target, the mode or rule that decided, and the remediation.

• Allow from the settings page — every recent interception offers an Allow action. It writes ONE minimal match: { network: { domains: [<host>] } } / action: allow rule at the top (index 0) of the nearest rule file that actually judges that connection: the project file of the workspace the block was attributed to (the absolute rulesFile when one is configured), or, for a session-less host-level block, the file the host chain resolves (absolute rulesFile → existing <processCwd>/<rulesFile> → configured fallbackPath → <processCwd>/<rulesFile>, created) — and when several workspaces are loaded, the page first offers a picker, because any loaded workspace chain outranks the host chain. Index 0 because rules are first-match-wins: appended after an existing deny the rule would be dead text. domains only, because that dimension is subdomain-inclusive and port/scheme-agnostic, so the host is not re-blocked on its next port; existing comments and untouched rules survive, and the text goes through the same validation gate as a hand edit. The change takes effect immediately — the cached workspace chains AND the session-less host chain are re-read, so no restart and no /rules reload — and the decision is recomputed, so the notice reports the REAL outcome and never claims success while the connection is still blocked. It refuses, leaving the file untouched, an unknown workspace, a target outside the known rule sources, the read-only built-in baseline, and a file it cannot read or parse; an already-allowed connection writes nothing. network.allowHostAction: false hides the button and makes the RPC refuse.

Quick start

# 1. install the bundle into your profile
dsh plugin --profile web add "github:PerryLink/dsh-permission-rules#main"

# or from npm (published releases)
dsh plugin --profile web add dsh-permission-rules

# 2. restart and verify the row
dsh --profile web --dump-config | grep -A4 'id: permission-rules'

Install & uninstall

• git channel (latest main): dsh plugin --profile web add "github:PerryLink/dsh-permission-rules#main" — the prepare script builds with production dependencies only.
• npm channel (published releases): dsh plugin --profile web add dsh-permission-rules.
• tarball channel: pnpm pack in this repo, then dsh plugin --profile web add ./dsh-permission-rules-<version>.tgz.
• uninstall: dsh plugin --profile web remove dsh-permission-rules.

Configuration

All tunables are Schemastery Config fields (changeable from cordis.yml). An id-targeted override replaces the whole row — restate every key you need. On a 0.1.7-alpha host every field is declared .volatile(), so it is ALSO editable in place from the Plugins settings page: the edit is committed into the running references without remounting the plugin, the runtime reads the new value on its next use, and a change to network.proxyPort, proxyBind, injectEnv or noProxy rebinds the local proxy immediately. The numeric bounds live in the schema, and the config editor resolves the merged config against that schema before persisting, so an out-of-range value is refused at write time. On a host whose Schemastery predates volatile() these fields stay cordis.yml-only.

Tools & surfaces

/rules                        list the active rules, their source files, and any last-reload error
/rules list                   explicit alias for the bare listing
/rules reload                 re-read the rule-file chain for this workspace
/rules decisions [n]          show the last n permission decisions of this session (default 10)
/rules test <tool> <json>     dry-evaluate the rules against a hypothetical call

/rules test also accepts leading flags: --cwd <dir>, --env KEY=VALUE (repeatable), --agent <selector> (repeatable), and --platform <name>. In multi-file chains (e.g. searchUp), every listed rule line is attributed to its own source file.

Permissions & data

• Permissions: declares files:read, files:watch, files:write, session:append, and network:outbound in its workshop manifest. ask decisions ride the official approval seam — nothing is re-implemented or bypassed.
• Data: rule files are read from disk; no rule data is written. No model calls, no reviewer subagents.
• Session log: permissionRules/decision is never injected into the model context and is appended with the envelope's ignorable: true marker so any harness build loads the log.

Security boundaries

• Policy, not a kernel. paths candidates come only from a documented set of argument keys (at any nesting depth, depth-capped), and only workspace-relative paths match.
• No reviewer here. The plugin never spawns subagents or calls models — producing an ask decision is the end of its work.
• No sandbox changes. OS-level sandbox policy belongs to the sandbox seam, not this plugin.
• Loud misconfiguration. Unknown YAML fields, unknown actions, and bad patterns are rejected at load.
• Backtracking bounds. Glob patterns are capped at maxGlobStars unbounded star expansions; regex-mode patterns reject nested unbounded quantifiers and quantified overlapping literal alternations.
• Fail-closed dialing. Plain-HTTP forwards and CONNECT tunnels connect only to adjudicated addresses; an allow whose adjudication resolved no address fails closed with 502 instead of dialing the name (issue #23).

Known limitations

• Audit marker on pre-marker and refusing hosts. permissionRules/decision is appended with ignorable: true; hosts whose Session.append predates the marker (the 0.1.0-rc.1–rc.7 and 0.1.1-rc.1–rc.7 lines) silently drop it, the 0.1.2-rc line ships the alpha.5 surface (no append option writes the marker), the 0.1.2-alpha line refuses plugin events on read even when marked, and the 0.1.3-alpha and 0.1.5-alpha lines keep the same surface-only append signature (verified on the published 0.1.3-alpha.1/0.1.5-alpha.1 packages) — the runtime pre-checks all of these before the first append and disables session-log audit with a one-time warning. Cross-generation migration refuses marked audit rows too: the 0.1.3-alpha v1→v2 gate refuses unknown v1 events, and the 0.1.5-alpha v2→v3 gate refuses every unclassified event (its inventory is frozen to the released v2 vocabulary), so strip v1 rows before a 0.1.3 host opens the log and v2 rows before upgrading to a 0.1.5 host. Native v3 logs accept marked plugin rows, so they only need repair. Set allowUnmarkedAudit: true to opt back in; repair already-written logs with scripts/repair-session-logs.mjs (its strip mode removes audit rows where the marker cannot help).
• Path candidates are heuristic. Only the documented argument keys feed path matching, and workspace-relative matching is ASCII-case-insensitive only when caseInsensitivePaths is on.
• Globs are a conservative subset. No brace expansion — write two patterns, or use regex mode.
• The regex backtracking guard is structural, not exhaustive. Prefer glob mode for untrusted files.
• The host's own outbound requests do not go through this proxy. The injected proxy environment covers spawned shell children — which is what the injection is for — and any consumer that reads the proxy names at request time. It does not cover the host process's own fetch-based traffic: the launcher installs undici's global dispatcher from the launch environment before the first plugin mounts, that dispatcher routes by its policy rather than by the environment, and Node samples the proxy environment at start — so a plugin mounting later cannot redirect it. Consequences: provider endpoints do not need an allow rule, and network.injectEnv: false does not take the host process out of anything. Measured 2026-09-10 on Node 22; reproduce with scripts/host-egress-probe.mjs.
• Upstream chaining covers only what reaches this proxy. On dsh-v0.1.7-alpha.1 and later, the harness's own launch-time proxy policy and its subprocess environment overlay can take precedence over the proxy names this plugin injects, so network.upstreamProxy describes what happens to the connections this proxy handles — it is not a claim about every outbound path in the host process.

Collaborating with dsh-auto-review

• dsh-permission-rules produces ask; dsh-auto-review answers on the approval/request waterfall with a read-only second-model verdict (or delegates to humans). Mount both for the full closed loop.
• Integration-tested: permissionRules/decision → approval/asked → autoReview/verdict → approval/decided, with the reviewer replaced by a scripted mock.
• The never approval policy and every fail-closed guarantee of the official harness stay untouched.

Session log repair

Session logs written before the ignorable marker existed can be refused by newer harness builds (SessionFormatUnsupportedError). The shipped scripts/repair-session-logs.mjs rewrites only the targeted audit rows to carry ignorable: true, frame-preserving, with backups:

node scripts/repair-session-logs.mjs scan [--home DIR]      # report foreign rows, change nothing
node scripts/repair-session-logs.mjs repair [--home DIR] [--dry-run]
node scripts/repair-session-logs.mjs strip [--home DIR] [--dry-run]

--home defaults to $DSH_HOME/sessions (or ~/.dsh/sessions). Every generation-addressed log is discovered by its canonical basename — session.jsonl, session.v2.jsonl, session.v3.jsonl, each optionally .zstd-compressed — so pick the mode by generation:

• v3 (session.v3.jsonl, written natively by the 0.1.5-alpha line) — the read path accepts marked plugin rows, so repair is enough.
• v2 (session.v2.jsonl, written by the 0.1.3-alpha line) — repair opens it on the host that wrote it, but the 0.1.5-alpha v2→v3 migration refuses every unclassified event even when marked: run strip on v2 logs before upgrading to a 0.1.5-alpha host.
• v1 (session.jsonl) — the 0.1.3-alpha v1→v2 migration refuses unknown v1 events even when marked: run strip before a 0.1.3-or-later host first opens the log.

Development

pnpm install            # node ^22.19 || >=24
pnpm run typecheck      # tsc, src + tests
pnpm run lint           # eslint, src + tests + scripts
pnpm test               # vitest: 24 spec files
pnpm run test:coverage  # coverage gate (90/80/90/90)
pnpm run build          # tsc declarations + tsdown bundles (lib/)
pnpm run pack:check     # build + pack (the published artifact)
node scripts/check-readme-sync.mjs   # five-language README sync gate (also in CI)

See VERIFICATION.md for the headless end-to-end verification record.

Interoperability with other DSH plugins

Verified against DSH 0.2.0-rc.2 (the runtime this README ships for) and the high-star plugin set surveyed on 2026-10-05.

This plugin does not interfere with other plugins, including the widely installed high-star ones:

• No tool-name collision. Every tool is namespaced; no bare name owned by a shipped tool or another plugin is registered.
• No service-key collision. It provides permissionRules, permissionRulesRuntime; those keys are not a built-in seam and are not provided by any surveyed high-star plugin.
• No slot collision. It registers no client slot key, so it cannot contend for a shadows-shipped-ui seat.
• No HTTP route collision. It registers no webServer prefix.
• No patch-layer collision. The bundle patch only inserts its own row; it never overrides a built-in row's config.
• No global mutation. It does not patch prototypes, rewrite process.env, or replace the global fetch dispatcher.

Shared event listeners are non-interfering by construction. It observes the ordering-sensitive events tools/post-execute, tools/pre-execute with ctx.on() — Cordis's broadcast registration, where every listener runs and none can starve another. Every listener here delegates through next(), so the chain is never short-circuited, and a mutation is applied to the value next() produced rather than returned in its place:

• tools/post-execute — also used by cc-safety-net (1576★).
• tools/pre-execute — also used by cc-safety-net (1576★).

Static evidence: dsh-plugin-doctor K10–K13 report pass for every check on this repository.

Topics

dsh, dsh-plugin, deepseek-harness, permission, policy, allow-deny-ask, approval, safety, network, network-policy, proxy

Contributors

• @PerryLink — creator and maintainer: rule vocabulary and evaluation, runtime, HMR watch, session-log audit, network policy + proxy, and the five-language docs.
• @22xuan — the detailed report on rc.6 hosts silently dropping the audit event's ignorable marker (#2) and the upstream harness discussion; the v0.4.1 runtime host-capability detection and the documentation correction drew directly from that analysis.
• @sjh9714 — proposed the shared rule-syntax test-vector corpus (#4, #5), shipped in v0.5.1 as docs/rule-test-vectors/, and supplied the AST-decomposition boundary cases on the design discussion.
• @weipeng1999 — the AST-based command-decomposition feature proposal (#8) behind the design discussion.
• @alexchenzl — the DSH Directory listing request (#7).
• @zl190 — reported and verified the 0.1.0-rc.7 harness compatibility gap (PR #9).
• @cuohua — reported that the 0.1.1-rc line still drops the ignorable marker even though the version gate covered only 0.1.0 (#11); the widened gate drew directly from that analysis.

PerryLink DSH Plugin Family

This project is one of the 44 DeepSeek Harness plugins maintained by PerryLink. If this one helps you, the others likely will too:

Install from the DSH Desktop Market

All PerryLink plugins are browsable in the built-in DSH Desktop Market: Market → Sources → add source → paste https://perrylink-dsh-catalog.perrylink.workers.dev/catalog-source.json → select it. Installation still goes through the Market's npm-identity verification and your confirmation.

License

Apache License 2.0 © 2026 dsh-permission-rules contributors