desktop: shell permission prompt has no description (missing settings.permissions.tool.shell i18n key) #59377

Activity

1. opencode-agent commented on Oct 26, 2026

   

   opencode-agentBot

   on Oct 26, 2026 – with OpenCode Agent

   Contributor

   More actions

   Thanks for the detailed report. I reproduced this on 2.1.4, and it still happens on the latest dev (ae876ee).

   With a { "action": "shell", "resource": "*", "effect": "ask" } rule and a model calling the shell tool on echo hello, the server's pending permission request has action: "shell" and resources: ["echo hello"]. The web/desktop permission dock looks up settings.permissions.tool.shell.description, but the app's translation files only define settings.permissions.tool.bash.* ("Run shell commands"). So the description line is never shown for shell permissions, in any locale. The script below reproduces the server side and prints the key the dock looks up.

   I couldn't render the desktop app itself here, and I didn't test the fully empty card for a request with no resources, since my test request always included the command. The TUI shows the shell prompt normally.

   Related: #51519 is the broader version of this, where MCP and other per-tool permissions get no hint at all.

   Reproduction script (drive.ts)

   // Repro for #59377: a shell permission request carries action "shell", but the
   // web/desktop permission dock (packages/app/src/session/requests/session-permission-dock.tsx)
   // looks up `settings.permissions.tool.<action>.description`, and the app's i18n bundle
   // only defines `settings.permissions.tool.bash.*` -> the description line is never shown.
   import { Effect, Stream } from "effect"
   import { Llm, OpenCodeDriver } from "opencode-drive"
   
   export default OpenCodeDriver.use(
     {
       ...(process.env.OC_DEV ? { opencode: { dev: process.env.OC_DEV } } : {}),
       tui: { recording: true },
       config: { permissions: [{ action: "shell", resource: "*", effect: "ask" }] } as any,
     },
     ({ ui, llm, opencode }) =>
       Effect.gen(function* () {
         let calls = 0
         yield* llm.serve((req: any) => {
           const system = String(req?.body?.messages?.[0]?.content ?? "")
           if (system.includes("You name conversations")) return Stream.make(Llm.text("Echo hello"))
           calls++
           return calls === 1
             ? Stream.make(
                 Llm.toolCall({ index: 0, id: "call_shell", name: "shell", input: { command: "echo hello" } }),
                 Llm.finish("tool-calls"),
               )
             : Stream.make(Llm.text("done"))
         })
         yield* ui.submit("Run echo hello")
         yield* Effect.sleep("5 seconds")
         yield* ui.screenshot("tui-permission")
         const pending: any = yield* opencode.permission.request.list()
         const actions = (pending.data ?? []).map((r: any) => r.action)
         console.log("PENDING PERMISSION REQUESTS:", JSON.stringify(pending.data, null, 2))
         console.log("ACTIONS:", actions.join(", "))
         console.log(
           "Desktop dock looks up:",
           actions.map((a: string) => `settings.permissions.tool.${a}.description`).join(", "),
         )
       }),
   )

   Run it with opencode-drive: opencode-drive run ./drive.ts. It drives opencode from your PATH; set OC_DEV to the path of an OpenCode checkout to run that from source instead.
2. opencode-agent

   added

   reproduced

   on Oct 26, 2026
3. pskraemer11 commented on Oct 26, 2026

   

   pskraemer11

   on Oct 26, 2026

   Author

   More actions

   Refinement after reading the source — resources is not empty for shell.

   packages/core/src/tool/plugin/shell.ts on dev:

   export const name = "shell"
   ...
   const parsed = yield* ShellParse.scan(invocation.command, invocation.shell, target.absolute, { portable })
   ...
   if (parsed.commands.length > 0)
     yield* permission.assert({
       action: name,
       resources: parsed.commands.map((command) => command.resource),
       save: parsed.commands.map((command) => command.save),
       ...

   So a shell permission always carries the parsed command(s) in resources, and the dock renders them as <code> lines. The body is therefore not empty — what is missing is only the description line above the command.

   This narrows the reported impact: the user can still see what is being run, they just get no hint line naming the action. A completely empty body, or one showing a bare *, must come from a different action — e.g. MCP tools, whose wrapper passes resources: ["*"]. That is #51519, not this issue.

   The fix is unchanged: add settings.permissions.tool.shell.title / .description to the i18n bundle, or map the existing bash entries onto shell in the lookup. The root cause — the bundle has bash while the server emits shell — is confirmed by the same file (export const name = "shell").
4. ZXLyyds100 commented on Oct 26, 2026

   

   ZXLyyds100

   on Oct 26, 2026

   More actions

   Implemented in #59405.

   • Added settings.permissions.tool.shell.title and .description, matching the existing shell wording.
   • Extended the existing permission-dock regression to assert the visible “Run shell commands” line.
   • Focused Chromium regression: 1 passed.
   • App unit command: 898 passed, 0 failed; app typecheck: 1,734 files, 0 errors.
   • Repository check: 35/35 typecheck tasks passed, 0 errors; existing lint warnings remain.

   The change is limited to the missing shell description; the parsed command resources remain visible, and MCP/other permission actions are unchanged. Real before/after captures from the same Chromium fixture are available as before and after.

   Implemented and verified with OpenAI Codex assistance. Hosted checks currently pass for standards, compliance, and contributor labeling; human review and any remaining CI are not implied.