fix permission config order (#24222)

Author: thdxr

packages/opencode/src/config/permission.ts

@@ -1,6 +1,7 @@
 export * as ConfigPermission from "./permission"
 import { Schema, SchemaGetter } from "effect"
-import { zod } from "@/util/effect-zod"
+import z from "zod"
+import { ZodOverride, zod } from "@/util/effect-zod"
 import { withStatics } from "@/util/schema"
 
 export const Action = Schema.Literals(["ask", "allow", "deny"])
@@ -18,17 +19,9 @@ export const Rule = Schema.Union([Action, Object])
   .pipe(withStatics((s) => ({ zod: zod(s) })))
 export type Rule = Schema.Schema.Type<typeof Rule>
 
-// Known permission keys get explicit types — most are full Rule (either a
-// single Action or a per-pattern object), but a handful of tools take no
-// sub-target patterns and are Action-only. Unknown keys fall through the
-// Record rest signature as Rule.
-//
-// StructWithRest canonicalises key order on decode (known first, then rest),
-// which used to require the `__originalKeys` preprocess hack because
-// `Permission.fromConfig` depended on the user's insertion order. That
-// dependency is gone — `fromConfig` now sorts top-level keys so wildcard
-// permissions come before specifics, making the final precedence
-// order-independent.
+// Known permission keys get explicit types in the Effect schema for generated
+// docs/types. Runtime config parsing uses `InfoZod` below so user key order is
+// preserved for permission precedence.
 const InputObject = Schema.StructWithRest(
   Schema.Struct({
     read: Schema.optional(Rule),
@@ -60,6 +53,18 @@ const InputSchema = Schema.Union([Action, InputObject])
 const normalizeInput = (input: Schema.Schema.Type<typeof InputSchema>): Schema.Schema.Type<typeof InputObject> =>
   typeof input === "string" ? { "*": input } : input
 
+const ACTION_ONLY = new Set(["todowrite", "question", "webfetch", "websearch", "codesearch", "doom_loop"])
+
+const InfoZod = z
+  .union([zod(Action), z.record(z.string(), z.union([zod(Action), z.record(z.string(), zod(Action))]))])
+  .transform(normalizeInput)
+  .superRefine((input, ctx) => {
+    for (const [key, value] of globalThis.Object.entries(input)) {
+      if (!ACTION_ONLY.has(key) || typeof value === "string") continue
+      ctx.addIssue({ code: "custom", message: `${key} must be a permission action`, path: [key] })
+    }
+  })
+
 export const Info = InputSchema.pipe(
   Schema.decodeTo(InputObject, {
     decode: SchemaGetter.transform(normalizeInput),
@@ -70,6 +75,7 @@ export const Info = InputSchema.pipe(
   }),
 )
   .annotate({ identifier: "PermissionConfig" })
+  .annotate({ [ZodOverride]: InfoZod })
   .pipe(
     // Walker already emits the decodeTo transform into the derived zod (see
     // `encoded()` in effect-zod.ts), so just expose that directly.

packages/opencode/src/permission/index.ts

@@ -288,18 +288,8 @@ function expand(pattern: string): string {
 }
 
 export function fromConfig(permission: ConfigPermission.Info) {
-  // Sort top-level keys so wildcard permissions (`*`, `mcp_*`) come before
-  // specific ones. Combined with `findLast` in evaluate(), this gives the
-  // intuitive semantic "specific tool rules override the `*` fallback"
-  // regardless of the user's JSON key order. Sub-pattern order inside a
-  // single permission key is preserved — only top-level keys are sorted.
-  const entries = Object.entries(permission).sort(([a], [b]) => {
-    const aWild = a.includes("*")
-    const bWild = b.includes("*")
-    return aWild === bWild ? 0 : aWild ? -1 : 1
-  })
   const ruleset: Ruleset = []
-  for (const [key, value] of entries) {
+  for (const [key, value] of Object.entries(permission)) {
     if (typeof value === "string") {
       ruleset.push({ permission: key, action: value, pattern: "*" })
       continue

packages/opencode/test/config/config.test.ts

@@ -1495,16 +1495,9 @@ test("merges legacy tools with existing permission config", async () => {
   })
 })
 
-test("permission config canonicalises known keys first, preserves rest-key insertion order", async () => {
-  // ConfigPermission.Info is a StructWithRest schema — the decoder reorders
-  // keys into declaration-order for known permission names (edit, read,
-  // todowrite, external_directory are declared in `config/permission.ts`),
-  // followed by rest keys in the user's insertion order.
-  //
-  // Rule precedence is NOT affected by this reordering: `Permission.fromConfig`
-  // sorts wildcards before specifics before iterating. See the
-  // "fromConfig - specific key beats wildcard regardless of JSON key order"
-  // test in test/permission/next.test.ts for the behavioural guarantee.
+test("permission config preserves user key order", async () => {
+  // Permission precedence follows the order users write in config, so parsing
+  // must not canonicalise known keys ahead of wildcard or custom keys.
   await using tmp = await tmpdir({
     init: async (dir) => {
       await Filesystem.write(
@@ -1532,15 +1525,12 @@ test("permission config canonicalises known keys first, preserves rest-key inser
     fn: async () => {
       const config = await load()
       expect(Object.keys(config.permission!)).toEqual([
-        // known fields that the user provided, in declaration order from
-        // config/permission.ts (read, edit, ..., external_directory, todowrite)
-        "read",
+        "*",
         "edit",
+        "write",
         "external_directory",
+        "read",
         "todowrite",
-        // rest keys (not in the known list), in user's insertion order
-        "*",
-        "write",
         "thoughts_*",
         "reasoning_model_*",
         "tools_*",

packages/opencode/test/permission/next.test.ts

@@ -128,61 +128,45 @@ test("fromConfig - does not expand tilde in middle of path", () => {
   expect(result).toEqual([{ permission: "external_directory", pattern: "/some/~/path", action: "allow" }])
 })
 
-// Top-level wildcard-vs-specific precedence semantics.
-//
-// fromConfig sorts top-level keys so wildcard permissions (containing "*")
-// come before specific permissions. Combined with `findLast` in evaluate(),
-// this gives the intuitive semantic "specific tool rules override the `*`
-// fallback", regardless of the order the user wrote the keys in their JSON.
-//
-// Sub-pattern order inside a single permission key (e.g. `bash: { "*": "allow", "rm": "deny" }`)
-// still depends on insertion order — only top-level keys are sorted.
-
-test("fromConfig - specific key beats wildcard regardless of JSON key order", () => {
+// Permission precedence follows config insertion order. `evaluate()` uses the
+// last matching rule, so later config entries intentionally override earlier
+// entries even when a wildcard appears after a specific permission.
+
+test("fromConfig - preserves top-level config key order", () => {
   const wildcardFirst = Permission.fromConfig({ "*": "deny", bash: "allow" })
   const specificFirst = Permission.fromConfig({ bash: "allow", "*": "deny" })
 
-  // Both orderings produce the same ruleset
-  expect(wildcardFirst).toEqual(specificFirst)
+  expect(wildcardFirst.map((r) => r.permission)).toEqual(["*", "bash"])
+  expect(specificFirst.map((r) => r.permission)).toEqual(["bash", "*"])
 
-  // And both evaluate bash → allow (bash rule wins over * fallback)
   expect(Permission.evaluate("bash", "ls", wildcardFirst).action).toBe("allow")
-  expect(Permission.evaluate("bash", "ls", specificFirst).action).toBe("allow")
+  expect(Permission.evaluate("bash", "ls", specificFirst).action).toBe("deny")
 })
 
-test("fromConfig - wildcard acts as fallback for permissions with no specific rule", () => {
-  const ruleset = Permission.fromConfig({ bash: "allow", "*": "ask" })
+test("fromConfig - wildcard acts as fallback when it appears before specifics", () => {
+  const ruleset = Permission.fromConfig({ "*": "ask", bash: "allow" })
   expect(Permission.evaluate("edit", "foo.ts", ruleset).action).toBe("ask")
   expect(Permission.evaluate("bash", "ls", ruleset).action).toBe("allow")
 })
 
-test("fromConfig - top-level ordering: wildcards first, specifics after", () => {
+test("fromConfig - top-level ordering is not sorted by wildcard specificity", () => {
   const ruleset = Permission.fromConfig({
     bash: "allow",
     "*": "ask",
     edit: "deny",
     "mcp_*": "allow",
   })
-  // wildcards (* and mcp_*) come before specifics (bash, edit)
-  const permissions = ruleset.map((r) => r.permission)
-  expect(permissions.slice(0, 2).sort()).toEqual(["*", "mcp_*"])
-  expect(permissions.slice(2)).toEqual(["bash", "edit"])
+  expect(ruleset.map((r) => r.permission)).toEqual(["bash", "*", "edit", "mcp_*"])
 })
 
-test("fromConfig - sub-pattern insertion order inside a tool key is preserved (only top-level sorts)", () => {
-  // Sub-patterns within a single tool key use the documented "`*` first,
-  // specific patterns after" convention (findLast picks specifics). The
-  // top-level sort must not touch sub-pattern ordering.
+test("fromConfig - sub-pattern insertion order inside a tool key is preserved", () => {
   const ruleset = Permission.fromConfig({ bash: { "*": "deny", "git *": "allow" } })
   expect(ruleset.map((r) => r.pattern)).toEqual(["*", "git *"])
-  // * fallback for unknown commands
   expect(Permission.evaluate("bash", "rm foo", ruleset).action).toBe("deny")
-  // specific pattern wins for git commands (it's last, findLast picks it)
   expect(Permission.evaluate("bash", "git status", ruleset).action).toBe("allow")
 })
 
-test("fromConfig - canonical documented example unchanged", () => {
-  // Regression guard for the example in docs/permissions.mdx
+test("fromConfig - documented fallback-first example", () => {
   const ruleset = Permission.fromConfig({ "*": "ask", bash: "allow", edit: "deny" })
   expect(Permission.evaluate("bash", "ls", ruleset).action).toBe("allow")
   expect(Permission.evaluate("edit", "foo.ts", ruleset).action).toBe("deny")
@@ -448,7 +432,7 @@ test("evaluate - wildcard permission fallback for unknown tool", () => {
   expect(result.action).toBe("ask")
 })
 
-test("evaluate - permission patterns sorted by length regardless of object order", () => {
+test("evaluate - later wildcard permission can override earlier specific permission", () => {
   const result = Permission.evaluate("bash", "rm", [
     { permission: "bash", pattern: "*", action: "allow" },
     { permission: "*", pattern: "*", action: "deny" },