feat: Add auto permission mode for Claude adapter

[charlesvien]

Problem

Claude upstream supports an auto mode that uses a model classifier to approve/deny permission prompts, but it was previously skipped in our fork.

Changes

1. Add auto to CODE_EXECUTION_MODES, available modes and AUTO_ALLOWED_TOOLS
2. Add "auto" option to ExitPlanMode permission options and approval handler
3. Map auto mode to Codex native auto mode
4. Update ModeSelector UI with Robot icon and blue styling for auto mode
5. Update UPSTREAM.md to reflect auto mode is now adopted

How did you test this?

Manually

[charlesvien]

• feat: Add usage limit detection and upgrade prompt #1750
• feat: Add auto permission mode for Claude adapter #1749 👈 (View in Graphite)
• feat: Sync Claude ACP adapter to v0.30.0 (SDK 0.2.112) #1748
• fix: Misc bugs and polish #1746
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[greptile-apps]

Comments Outside Diff (2)

1. packages/agent/src/execution-mode.test.ts, line 5-22 (link)

   Tests assume non-root environment

   Both tests hard-code bypassPermissions and full-access in their expected arrays, but those modes are only pushed onto the lists when ALLOW_BYPASS = !IS_ROOT. If the test process runs as UID 0 (common in Docker-based CI), IS_ROOT is true, neither mode is ever added, and the actual arrays will be ["auto","default","acceptEdits","plan"] / ["read-only","auto"] — causing both tests to fail without any obvious connection to the root check.

   The tests need either a jest/vitest mock for IS_ROOT, a vi.mock("./utils/common", ...) call, or separate test cases for each branch.

   Prompt To Fix With AI

   This is a comment left during a code review.
   Path: packages/agent/src/execution-mode.test.ts
   Line: 5-22
   
   Comment:
   **Tests assume non-root environment**
   
   Both tests hard-code `bypassPermissions` and `full-access` in their expected arrays, but those modes are only pushed onto the lists when `ALLOW_BYPASS = !IS_ROOT`. If the test process runs as UID 0 (common in Docker-based CI), `IS_ROOT` is `true`, neither mode is ever added, and the actual arrays will be `["auto","default","acceptEdits","plan"]` / `["read-only","auto"]` — causing both tests to fail without any obvious connection to the root check.
   
   The tests need either a jest/vitest mock for `IS_ROOT`, a `vi.mock("./utils/common", ...)` call, or separate test cases for each branch.
   
   How can I resolve this? If you propose a fix, please make it concise.

2. packages/agent/src/execution-mode.test.ts, line 1-22 (link)

   Prefer parameterised tests

   Both test cases call getModes().map((mode) => mode.id) and compare against a fixed array — the same assertion shape for two different adapters. Per the team's convention, these should use it.each:

   it.each([
     [
       "claude sessions",
       getAvailableModes,
       ["auto", "default", "acceptEdits", "plan", "bypassPermissions"],
     ],
     ["codex sessions", getAvailableCodexModes, ["read-only", "auto", "full-access"]],
   ])(
     "includes correct modes for %s",
     (_label, getModes, expected) => {
       expect(getModes().map((mode) => mode.id)).toEqual(expected);
     },
   );

   Prompt To Fix With AI

   This is a comment left during a code review.
   Path: packages/agent/src/execution-mode.test.ts
   Line: 1-22
   
   Comment:
   **Prefer parameterised tests**
   
   Both test cases call `getModes().map((mode) => mode.id)` and compare against a fixed array — the same assertion shape for two different adapters. Per the team's convention, these should use `it.each`:
   
   ```typescript
   it.each([
     [
       "claude sessions",
       getAvailableModes,
       ["auto", "default", "acceptEdits", "plan", "bypassPermissions"],
     ],
     ["codex sessions", getAvailableCodexModes, ["read-only", "auto", "full-access"]],
   ])(
     "includes correct modes for %s",
     (_label, getModes, expected) => {
       expect(getModes().map((mode) => mode.id)).toEqual(expected);
     },
   );
   ```
   
   How can I resolve this? If you propose a fix, please make it concise.

   Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

Prompt To Fix All With AI

This is a comment left during a code review.
Path: packages/agent/src/execution-mode.test.ts
Line: 5-22

Comment:
**Tests assume non-root environment**

Both tests hard-code `bypassPermissions` and `full-access` in their expected arrays, but those modes are only pushed onto the lists when `ALLOW_BYPASS = !IS_ROOT`. If the test process runs as UID 0 (common in Docker-based CI), `IS_ROOT` is `true`, neither mode is ever added, and the actual arrays will be `["auto","default","acceptEdits","plan"]` / `["read-only","auto"]` — causing both tests to fail without any obvious connection to the root check.

The tests need either a jest/vitest mock for `IS_ROOT`, a `vi.mock("./utils/common", ...)` call, or separate test cases for each branch.

How can I resolve this? If you propose a fix, please make it concise.

---

This is a comment left during a code review.
Path: packages/agent/src/execution-mode.test.ts
Line: 1-22

Comment:
**Prefer parameterised tests**

Both test cases call `getModes().map((mode) => mode.id)` and compare against a fixed array — the same assertion shape for two different adapters. Per the team's convention, these should use `it.each`:

```typescript
it.each([
  [
    "claude sessions",
    getAvailableModes,
    ["auto", "default", "acceptEdits", "plan", "bypassPermissions"],
  ],
  ["codex sessions", getAvailableCodexModes, ["read-only", "auto", "full-access"]],
])(
  "includes correct modes for %s",
  (_label, getModes, expected) => {
    expect(getModes().map((mode) => mode.id)).toEqual(expected);
  },
);
```

How can I resolve this? If you propose a fix, please make it concise.

_{Reviews (1): Last reviewed commit: "Add auto permission mode for Claude adap..." | Re-trigger Greptile}

[jonathanlab]

Super excited about this!

[charlesvien]

Merge activity

• Apr 21, 8:05 PM UTC: @charlesvien merged this pull request with Graphite.