Droid agent (#1318)

* droid research (vibe-kanban 054135e9)

<droid-docs>
# Overview

> Non-interactive execution mode for CI/CD pipelines and automation scripts.

# Droid Exec (Headless CLI)

Droid Exec is Factory's headless execution mode designed for automation workflows. Unlike the interactive CLI, `droid exec` runs as a one-shot command that completes a task and exits, making it ideal for CI/CD pipelines, shell scripts, and batch processing.

## Summary and goals

Droid Exec is a one-shot task runner designed to:

* Produce readable logs, and structured artifacts when requested
* Enforce opt-in for mutations/command execution (secure-by-default)
* Fail fast on permission violations with clear errors
* Support simple composition for batch and parallel work

<CardGroup cols={2}>
  <Card title="Non-Interactive" icon="terminal">
    Single run execution that writes to stdout/stderr for CI/CD integration
  </Card>

  <Card title="Secure by Default" icon="lock">
    Read-only by default with explicit opt-in for mutations via autonomy levels
  </Card>

  <Card title="Composable" icon="puzzle">
    Designed for shell scripting, parallel execution, and pipeline integration
  </Card>

  <Card title="Clean Output" icon="file-export">
    Structured output formats and artifacts for automated processing
  </Card>
</CardGroup>

## Execution model

* Non-interactive single run that writes to stdout/stderr.
* Default is spec-mode: the agent is only allowed to execute read-only operations.
* Add `--auto` to enable edits and commands; risk tiers gate what can run.

CLI help (excerpt):

```
Usage: droid exec [options] [prompt]

Execute a single command (non-interactive mode)

Arguments:
  prompt                          The prompt to execute

Options:
  -o, --output-format <format>    Output format (default: "text")
  -f, --file <path>               Read prompt from file
  --auto <level>                  Autonomy level: low|medium|high
  --skip-permissions-unsafe       Skip ALL permission checks (unsafe)
  -s, --session-id <id>           Existing session to continue (requires a prompt)
  -m, --model <id>                Model ID to use
  -r, --reasoning-effort <level>  Reasoning effort: off|low|medium|high
  --cwd <path>                    Working directory path
  -h, --help                      display help for command
```

Supported models (examples):

* gpt-5-codex (default)
* gpt-5-2025-08-07
* claude-sonnet-4-20250514
* claude-opus-4-1-20250805

## Installation

<Steps>
  <Step title="Install Droid CLI">
    <CodeGroup>
      ```bash macOS/Linux theme={null}
      curl -fsSL https://app.factory.ai/cli | sh
      ```

      ```powershell Windows theme={null}
      irm https://app.factory.ai/cli/windows | iex
      ```
    </CodeGroup>
  </Step>

  <Step title="Get Factory API Key">
    Generate your API key from the [Factory Settings Page](https://app.factory.ai/settings/api-keys)
  </Step>

  <Step title="Set Environment Variable">
    Export your API key as an environment variable:

    ```bash  theme={null}
    export FACTORY_API_KEY=fk-...
    ```
  </Step>
</Steps>

## Quickstart

* Direct prompt:
  * `droid exec "analyze code quality"`
  * `droid exec "fix the bug in src/main.js" --auto low`
* From file:
  * `droid exec -f prompt.md`
* Pipe:
  * `echo "summarize repo structure" | droid exec`
* Session continuation:
  * `droid exec --session-id <session-id> "continue with next steps"`

## Autonomy Levels

Droid exec uses a tiered autonomy system to control what operations the agent can perform. By default, it runs in read-only mode, requiring explicit flags to enable modifications.

### DEFAULT (no flags) - Read-only Mode

The safest mode for reviewing planned changes without execution:

* ✅ Reading files or logs: cat, less, head, tail, systemctl status
* ✅ Display commands: echo, pwd
* ✅ Information gathering: whoami, date, uname, ps, top
* ✅ Git read operations: git status, git log, git diff
* ✅ Directory listing: ls, find (without -delete or -exec)
* ❌ No modifications to files or system
* **Use case:** Safe for reviewing what changes would be made

```bash  theme={null}
# Analyze and plan refactoring without making changes
droid exec "Analyze the authentication system and create a detailed plan for migrating from session-based auth to OAuth2. List all files that would need changes and describe the modifications required."

# Review code quality and generate report
droid exec "Review the codebase for security vulnerabilities, performance issues, and code smells. Generate a prioritized list of improvements needed."

# Understand project structure
droid exec "Analyze the project architecture and create a dependency graph showing how modules interact with each other."
```

### `--auto low` - Low-risk Operations

Enables basic file operations while blocking system changes:

* ✅ File creation/editing in project directories
* ❌ No system modifications or package installations
* **Use case:** Documentation updates, code formatting, adding comments

```bash  theme={null}
# Safe file operations
droid exec --auto low "add JSDoc comments to all functions"
droid exec --auto low "fix typos in README.md"
```

### `--auto medium` - Development Operations

Operations that may have significant side effects, but these side effects are typically harmless and straightforward to recover from.
Adds common development tasks to low-risk operations:

* Installing packages from trusted sources: npm install, pip install (without sudo)
* Network requests to trusted endpoints: curl, wget to known APIs
* Git operations that modify local repositories: git commit, git checkout, git pull (but not git push)
* Building code with tools like make, npm run build, mvn compile
* ❌ No git push, sudo commands, or production changes
* **Use case:** Local development, testing, dependency management

```bash  theme={null}
# Development tasks
droid exec --auto medium "install deps, run tests, fix issues"
droid exec --auto medium "update packages and resolve conflicts"
```

### `--auto high` - Production Operations

Commands that may have security implications such as data transfers between untrusted sources or execution of unknown code, or major side effects such as irreversible data loss or modifications of production systems/deployments.

* Running arbitrary/untrusted code: curl | bash, eval, executing downloaded scripts
* Exposing ports or modifying firewall rules that could allow external access
* Git push operations that modify remote repositories: git push, git push --force
* Irreversible actions to production deployments, database migrations, or other sensitive operations
* Commands that access or modify sensitive information like passwords or keys
* ❌ Still blocks: sudo rm -rf /, system-wide changes
* **Use case:** CI/CD pipelines, automated deployments

```bash  theme={null}
# Full workflow automation
droid exec --auto high "fix bug, test, commit, and push to main"
droid exec --auto high "deploy to staging after running tests"
```

### `--skip-permissions-unsafe` - Bypass All Checks

<Warning>
  DANGEROUS: This mode allows ALL operations without confirmation. Only use in completely isolated environments like Docker containers or throwaway VMs.
</Warning>

* ⚠️ Allows ALL operations without confirmation
* ⚠️ Can execute irreversible operations
* Cannot be combined with --auto flags
* **Use case:** Isolated environments

```bash  theme={null}
# In a disposable Docker container for CI testing
docker run --rm -v $(pwd):/workspace alpine:latest sh -c "
  apk add curl bash &&
  curl -fsSL https://app.factory.ai/cli | sh &&
  droid exec --skip-permissions-unsafe 'Install all system dependencies, modify system configs, run integration tests that require root access, and clean up test databases'
"

# In ephemeral GitHub Actions runner for rapid iteration
# where the runner is destroyed after each job
droid exec --skip-permissions-unsafe "Modify /etc/hosts for test domains, install custom kernel modules, run privileged container tests, and reset network interfaces"

# In a temporary VM for security testing
droid exec --skip-permissions-unsafe "Run penetration testing tools, modify firewall rules, test privilege escalation scenarios, and generate security audit reports"
```

### Fail-fast Behavior

If a requested action exceeds the current autonomy level, droid exec will:

1. Stop immediately with a clear error message
2. Return a non-zero exit code
3. Not perform any partial changes

This ensures predictable behavior in automation scripts and CI/CD pipelines.

## Output formats and artifacts

Droid exec supports three output formats for different use cases:

### text (default)

Human-readable output for direct consumption or logs:

```bash  theme={null}
$ droid exec --auto low "create a python file that prints 'hello world'"
Perfect! I've created a Python file named `hello_world.py` in your home directory that prints 'hello world' when executed.
```

### json

Structured JSON output for parsing in scripts and automation:

```bash  theme={null}
$ droid exec "summarize this repository" --output-format json
{
  "type": "result",
  "subtype": "success",
  "is_error": false,
  "duration_ms": 5657,
  "num_turns": 1,
  "result": "This is a Factory documentation repository containing guides for CLI tools, web platform features, and onboarding procedures...",
  "session_id": "8af22e0a-d222-42c6-8c7e-7a059e391b0b"
}
```

Use JSON format when you need to:

* Parse the result in a script
* Check success/failure programmatically
* Extract session IDs for continuation
* Process results in a pipeline

### debug

Streaming messages showing the agent's execution in real-time:

```bash  theme={null}
$ droid exec "run ls command" --output-format debug
{"type":"message","role":"user","text":"run ls command"}
{"type":"message","role":"assistant","text":"I'll run the ls command to list the contents..."}
{"type":"tool_call","toolName":"Execute","parameters":{"command":"ls -la"}}
{"type":"tool_result","value":"total 16\ndrwxr-xr-x@ 8 user staff..."}
{"type":"message","role":"assistant","text":"The ls command has been executed successfully..."}
```

Debug format is useful for:

* Monitoring agent behavior
* Troubleshooting execution issues
* Understanding tool usage patterns
* Real-time progress tracking

For automated pipelines, you can also direct the agent to write specific artifacts:

```bash  theme={null}
droid exec --auto low "Analyze dependencies and write to deps.json"
droid exec --auto low "Generate metrics report in CSV format to metrics.csv"
```

## Working directory

* Use `--cwd` to scope execution:

```
droid exec --cwd /home/runner/work/repo "Map internal packages and dump graphviz DOT to deps.dot"
```

## Models and reasoning effort

* Choose a model with `-m` and adjust reasoning with `-r`:

```
droid exec -m claude-sonnet-4-20250514 -r medium -f plan.md
```

## Batch and parallel patterns

Shell loops (bounded concurrency):

```bash  theme={null}
# Process files in parallel (GNU xargs -P)
find src -name "*.ts" -print0 | xargs -0 -P 4 -I {} \
  droid exec --auto low "Refactor file: {} to use modern TS patterns"
```

Background job parallelization:

```bash  theme={null}
# Process multiple directories in parallel with job control
for path in packages/ui packages/models apps/factory-app; do
  (
    cd "$path" &&
    droid exec --auto low "Run targeted analysis and write report.md"
  ) &
done
wait  # Wait for all background jobs to complete
```

Chunked inputs:

```bash  theme={null}
# Split large file lists into manageable chunks
git diff --name-only origin/main...HEAD | split -l 50 - /tmp/files_
for f in /tmp/files_*; do
  list=$(tr '\n' ' ' < "$f")
  droid exec --auto low "Review changed files: $list and write to review.json"
done
rm /tmp/files_*  # Clean up temporary files
```

Workflow Automation (CI/CD):

```yaml  theme={null}
# Dead code detection and cleanup suggestions
name: Code Cleanup Analysis
on:
  schedule:
    - cron: '0 1 * * 0' # Weekly on Sundays
  workflow_dispatch:
jobs:
  cleanup-analysis:
    strategy:
      matrix:
        module: ['src/components', 'src/services', 'src/utils', 'src/hooks']
    steps:
      - uses: actions/checkout@v4
      - run: droid exec --cwd "${{ matrix.module }}" --auto low "Identify unused exports, dead code, and deprecated patterns. Generate cleanup recommendations in cleanup-report.md"
```

## Unique usage examples

License header enforcer:

```bash  theme={null}
git ls-files "*.ts" | xargs -I {} \
  droid exec --auto low "Ensure {} begins with the Apache-2.0 header; add it if missing"
```

API contract drift check (read-only):

```bash  theme={null}
droid exec "Compare openapi.yaml operations to our TypeScript client methods and write drift.md with any mismatches"
```

Security sweep:

```bash  theme={null}
droid exec --auto low "Run a quick audit for sync child_process usage and propose fixes; write findings to sec-audit.csv"
```

## Exit behavior

* 0: success
* Non-zero: failure (permission violation, tool error, unmet objective). Treat non-zero as failed in CI.

## Best practices

* Favor `--auto low`; keep mutations minimal and commit/push in scripted steps.
* Avoid `--skip-permissions-unsafe` unless fully sandboxed.
* Ask the agent to emit artifacts your pipeline can verify.
* Use `--cwd` to constrain scope in monorepos.

</droid-docs>

Use the oracle to research how we support custom executors.
AMP and Claude Code would likely be good references here as I believe that they both operate via JSON.

Save your findings in a single markdown file.

* begin droid

* add plan

* droid implementation (vibe-kanban 90e6c8f6)

Read tasks/droid-agent/plan.md and execute the plan.

* document droid (vibe-kanban 0a7f8590)

we have introduced a new coding agent

Installation instructions are at https://factory.ai/product/cli

We expect that users have the `droid` cli installed and that they have logged in.

docs/supported-coding-agents.mdx
There may also be other docs or references.

* red gh action (vibe-kanban f0c8b6c4)

Run cargo fmt --all -- --check
  cargo fmt --all -- --check
  npm run generate-types:check
  cargo test --workspace
  cargo clippy --all --all-targets -- -D warnings

the checks step is failing, can you see what's up with the rust codebase and resolve it?

* droid | settings bug (vibe-kanban 7deec8df)

We have a new coding agent called Droid and it has a variety of different settings including the autonomy level and we default this to medium and users can update this by going to settings and then using the drop down to change it and then hitting the save button. And this works, however, when users return back to settings the displayed autonomy level is reset to medium rather than the correct level. So can you investigate why this is happening and plan how we can improve it, how we can verify it, do we need to introduce some logging, other things to consider. Write up your plan in a new markdown file.

* glob

* tool call parsing & display (vibe-kanban e3f65a74)

droid.rs has `fn map_tool_to_action`

The problem is that we're doing a poor job at displaying these tool calls e.g. glob. In `claude.rs`, we use `ClaudeToolData`, a struct that matches the real JSON data. Once we do that, we have a type safe way to map tool calls to the `ActionType` struct.

You can run `droid exec --output-format=stream-json --auto medium "YOUR MESSAGE MERE"` in a temporary directory to instruct the agent to generate custom outputs in case you need more sample data.

I just added glob.jsonl under droid-json, there are other json files in there too.

I recommend using sub agents as some of these files (e.g. claude.rs) are large.

cursor.rs might also be a useful reference.

You're done once we properly handle these tools.

* show droid model (vibe-kanban 8fdbc630)

The first JSON object emitted from the droid executor is a system message with a `model` field. We should capture and display this.

I believe that we're already doing something similar with Codex.

Here's a sample system message:
{"type":"system","subtype":"init","cwd":"/Users/britannio/projects/vibe-kanban","session_id":"59a75629-c0c4-451f-a3c7-8e9eab05484a","tools":["Read","LS","Execute","Edit","MultiEdit","ApplyPatch","Grep","Glob","Create","ExitSpecMode","WebSearch","TodoWrite","FetchUrl","slack_post_message"],"model":"gpt-5-codex"}

* reliable apply patch display (vibe-kanban 3710fb65)

The crates/executors/src/executors/droid.rs ApplyPatch tool call contains an `input` string which isn't very helpful, but the tool call result is a JSON object with a `value` object with the fields success, content, diff, and file_path.

Here's a parsed example of `value`:
{
  "success": true,
  "content": "def bubble_sort(arr):\n    \"\"\"\n    Bubble Sort Algorithm\n    Time Complexity: O(n^2)\n    Space Complexity: O(1)\n\n    Repeatedly steps through the list, compares adjacent elements and swaps them\n    if they are in the wrong order.\n    \"\"\"\n    n = len(arr)\n    arr = arr.copy()  # Create a copy to avoid modifying the original\n\n    for i in range(n):\n        # Flag to optimize by stopping if no swaps occur\n        swapped = False\n\n        for j in range(0, n - i - 1):\n            if arr[j] > arr[j + 1]:\n                arr[j], arr[j + 1] = arr[j + 1], arr[j]\n                swapped = True\n\n        # If no swaps occurred, array is already sorted\n        if not swapped:\n            break\n\n    return arr\n\n\ndef insertion_sort(arr):\n    \"\"\"\n    Insertion Sort Algorithm\n    Time Complexity: O(n^2)\n    Space Complexity: O(1)\n\n    Builds the sorted portion of the array one element at a time by inserting\n    each element into its correct position.\n    \"\"\"\n    arr = arr.copy()  # Create a copy to avoid modifying the original\n\n    for i in range(1, len(arr)):\n        key = arr[i]\n        j = i - 1\n\n        while j >= 0 and arr[j] > key:\n            arr[j + 1] = arr[j]\n            j -= 1\n\n        arr[j + 1] = key\n\n    return arr\n\n\nif __name__ == \"__main__\":\n    # Example usage\n    test_array = [64, 34, 25, 12, 22, 11, 90]\n\n    print(\"Original array:\", test_array)\n    print(\"\\nBubble Sort result:\", bubble_sort(test_array))\n    print(\"Insertion Sort result:\", insertion_sort(test_array))\n\n    # Test with different arrays\n    print(\"\\n--- Additional Tests ---\")\n    test_cases = {\n        \"Reverse sorted\": [5, 4, 3, 2, 1],\n        \"Empty array\": [],\n        \"Already sorted\": [1, 2, 3, 4, 5],\n    }\n\n    for description, case in test_cases.items():\n        print(f\"{description} (Bubble):\", bubble_sort(case))\n        print(f\"{description} (Insertion):\", insertion_sort(case))\n",
  "diff": "--- previous\t\n+++ current\t\n@@ -26,14 +26,46 @@\n     return arr\n \n \n+def insertion_sort(arr):\n+    \"\"\"\n+    Insertion Sort Algorithm\n+    Time Complexity: O(n^2)\n+    Space Complexity: O(1)\n+\n+    Builds the sorted portion of the array one element at a time by inserting\n+    each element into its correct position.\n+    \"\"\"\n+    arr = arr.copy()  # Create a copy to avoid modifying the original\n+\n+    for i in range(1, len(arr)):\n+        key = arr[i]\n+        j = i - 1\n+\n+        while j >= 0 and arr[j] > key:\n+            arr[j + 1] = arr[j]\n+            j -= 1\n+\n+        arr[j + 1] = key\n+\n+    return arr\n+\n+\n if __name__ == \"__main__\":\n     # Example usage\n     test_array = [64, 34, 25, 12, 22, 11, 90]\n \n     print(\"Original array:\", test_array)\n     print(\"\\nBubble Sort result:\", bubble_sort(test_array))\n+    print(\"Insertion Sort result:\", insertion_sort(test_array))\n \n     # Test with different arrays\n     print(\"\\n--- Additional Tests ---\")\n-    print(\"Reverse sorted:\", bubble_sort([5, 4, 3, 2, 1]))\n-    print(\"Empty array:\", bubble_sort([]))\n+    test_cases = {\n+        \"Reverse sorted\": [5, 4, 3, 2, 1],\n+        \"Empty array\": [],\n+        \"Already sorted\": [1, 2, 3, 4, 5],\n+    }\n+\n+    for description, case in test_cases.items():\n+        print(f\"{description} (Bubble):\", bubble_sort(case))\n+        print(f\"{description} (Insertion):\", insertion_sort(case))",
  "file_path": "/Users/britannio/projects/droid-simple/sorting_algorithms.py"
}

This formatting should be deterministic and thus we can use it to show more informative tool call data.

The first thing to understand is if this will naturally fit with the current architecture, as we only reliably know how the file has changed (and what the target file was) after receiving the tool call result.

* droid failed tool call handling (vibe-kanban bd7feddb)

crates/executors/src/executors/droid.rs
droid-json/insufficient-perms.jsonl

the insufficient-perms file contains the JSON output log of a run where it runs a command to create a file but the tool call fails due to a permission error.

I'd expect that the failed tool result would be correlated with the tool call and thus i'd see an ARGS block and a RESULTS block within the tool call on the front-end.

Instead, I see the tool call only with the ARGS block, then I see a separate UI element with the JSON tool result as if it failed to be correlated.

Firstly, I want to follow TDD by creating a failing test that confirms this behaviour. It might be hard though because we haven't designed the code in droid.rs with testability in mind.

Lets first analyse the code to consider if it's already testable or if we need to do any refactoring & introduce harnesses etc.

My perspective of the coding agent is that we send it a command, and it streams JSON objects one by one so some form of reducer pattern seems natural (previous list of json objects + previous state + new json object => new state). Either 'new state' or 'new delta'.

When we resume a session, it will emit a system message object, then a message object with role user (repeating what we sent it), then the new actions that it takes.

* droid default (vibe-kanban 2f8a19cc)

the default autonomy level is currently medium. Lets change it to the highest (unsafe)

* droid globbing rendering (vibe-kanban 76d372ea)

See droid-json/glob.jsonl
Notice the `patterns` field. Unfortunately, we seems to not be using this data as glob tool calls are being rendered exclusively via a file name of some sort rather than `Globbing README.md, readme.md,docs/**,*.md`

Use the oracle to investigate this.

* droid todo list text (vibe-kanban b1bdeffc)

Use the text 'TODO list updated' for the droid agent when it makes a change to the todo list.

* droid workspace path (vibe-kanban 0486b74a)

See how claude.rs uses worktree_path (from normalize_logs).
We should be doing the same for the droid executor so that the tool calls we generate have relative paths.

* mcp settings (vibe-kanban 2031d8f4)

Quick fix: Filter that agent from the dropdown in the frontend.

// In McpSettings.tsx, line 282-289
<SelectContent>
  {profiles &&
    Object.entries(profiles)
      .filter(([key]) => key !== 'DROID') // or whatever the agent name is
      .sort((a, b) => a[0].localeCompare(b[0]))
      .map(([profileKey]) => (
        <SelectItem key={profileKey} value={profileKey}>
          {profileKey}
        </SelectItem>
      ))}
</SelectContent>

we need to temporarily hide droid as it doesn't support mcp yet.

* clean up (vibe-kanban 6b1a8e2e)

remove all references to 'britannio' from the droid module.

* delete droid json

* droid agent code review (vibe-kanban 6820ffd1)

We added Droid to crates/services/src/services/config/versions/v1.rs but presumably we should've used the latest reasonable version. See what we used for Copilot.

Delete docs/adr-droid-architecture.md
Delete docs/droid-improvements-summary.md

docs/supported-coding-agents.mdx the default was medium, it's now skip-permissions-unsafe

Delete the tasks/ folder

* remove unnecessary v1 change

* updated droid.json schema

* tweak command

* droid model suggestions (vibe-kanban 120f87d2)

crates/executors/src/executors/droid/types.rs

Valid model IDs are:
  gpt-5-codex                  OpenAI GPT-5-Codex (Auto)
  claude-sonnet-4-5-20250929   Claude Sonnet 4.5
  gpt-5-2025-08-07             OpenAI GPT-5
  claude-opus-4-1-20250805     Claude Opus 4.1
  claude-haiku-4-5-20251001    Claude Haiku 4.5
  glm-4.6                      Droid Core (GLM-4.6)

We currently mention gpt-5-codex, claude-sonnet-4

* remove dead code

* droid automated testing (vibe-kanban f836b4a4)

lets start brainstorming this, starting with tests in crates/executors/src/executors/droid/types.rs to ensure that we correctly generate a command

* create exec_command_with_prompt

* Add logging to error paths in action_mapper.rs (vibe-kanban 76cc5d71)

Add tracing logging (warn/error) to error paths in `crates/executors/src/executors/droid/action_mapper.rs` following existing logging patterns in the codebase.

Key locations:
- Line 32-35: DroidToolData parsing failure (currently silent)
- Any other error paths that swallow errors

Use `tracing::warn!` with structured fields for context (tool_name, error details, etc.)

* droid automated testing (DroidJSON -> NormalizedEntry) (vibe-kanban cf325d24)

We have example agent from /Users/britannio/Downloads/droid-json

Read crates/executors/src/executors/droid/events.rs

Use the oracle to plan tests that we could introduce.

* preserve timestamp

* droid reasoning effort (vibe-kanban 47dae2db)

in settings, we're showing a dropdown for the droid autonomy level. We should be doing the same for the reasoning level. It should default to being empty if possible.

* droid path (vibe-kanban d8370535)

Droid file edits (presumably ApplyPatch?) aren't using relative paths. E.g. i'm seeing `/private/var/folders/5q/5vgq75y92dz0k7n62z93299r0000gn/T/vibe-kanban-dev/worktrees/11dc-setup/next.config.mjs`

* fix warning

* fix warning

* whitespace update

* DomainEvent -> LogEvent

* remove msg store stream -> line converter

* normalise the diff generated when the droid ApplyPatch tool call is
parsed

* refactor process_event to mutate a reference to ProcessorState

* remove EntryIndexProvider abstraction

* remove dead code

* remove JSON indirection when invoking extract_path_from_patch

* converting DroidJson -> LogEvent produces Option instead of Vec
DroidJson mapping tests removed in favour of snapshot testing delete
emit_patches (now redundant) update match syntax in
compute_updated_action_type make process_event a member of
ProcessorState

* simplify droid build_command_builder

* simplify droid types tests

* remove droid type tests

* rename events.rs -> log_event_converter.rs
rename patch_emitter -> patch_converter
remove ParsedLine indirection from processor.rs
handle Edit, MultiEdit, and Create tool calls (only used by some models like claude)
move action mapper logic to log_event_converter
introduce a claude snapshot
update snapshots

* add error log for failed parsing of DroidJson

* update snapshots

* Fix clippy warnings in droid executor

- Change &String to &str in extract_path_from_patch
- Rename to_patch to process_event for correct self convention

Amp-Thread-ID: https://ampcode.com/threads/T-81d4f5ac-6d3a-4da5-9799-de724f3df1e3
Co-authored-by: Amp <amp@ampcode.com>

* update cargo lock

* droid tool call result parsing (vibe-kanban 514d27de)

the droid executor has a regression where the `droid exec` command is no longer producing an `id` field for tool_result messages. Fortunately, in most cases, it's safe to stick to FIFO behaviour whereby if we get a tool result, we can match it with the earliest tool call. This won't always work but it's a reasonable solution for the next few days while the droid team fixes their executor.

Start by using the oracle to trace and understand the codepaths involved, and to make a plan. We likely need to update the DroidJson struct so that the tool call result id becomes optional.

To test this, we can take an existing snapshot test and create a variant of it without ids in the tool call results, and see if we still produce equivalent log events.

* refactor: collapse nested if statements in log_event_converter

Amp-Thread-ID: https://ampcode.com/threads/T-b9ad8aac-0fd5-44c5-b2f8-317d79b623a6
Co-authored-by: Amp <amp@ampcode.com>

* format

* Cleanup droid executor implementation

* Implement session forking

* linter

---------

Co-authored-by: Britannio Jarrett <britanniojarrett@gmail.com>
Co-authored-by: Test User <test@example.com>
Co-authored-by: Amp <amp@ampcode.com>

crates/executors/default_profiles.json

@@ -134,6 +134,55 @@
           "model": "claude-sonnet-4"
         }
       }
+    },
+    "DROID": {
+      "DEFAULT": {
+        "DROID": {
+          "autonomy": "skip-permissions-unsafe"
+        }
+      },
+      "GPT_5_1_CODEX": {
+        "DROID": {
+          "autonomy": "skip-permissions-unsafe",
+          "model": "gpt-5.1-codex"
+        }
+      },
+      "GPT_5_1": {
+        "DROID": {
+          "autonomy": "skip-permissions-unsafe",
+          "model": "gpt-5.1"
+        }
+      },
+      "GEMINI_3_PRO": {
+        "DROID": {
+          "autonomy": "skip-permissions-unsafe",
+          "model": "gemini-3-pro-preview"
+        }
+      },
+      "CLAUDE_SONNET_4_5": {
+        "DROID": {
+          "autonomy": "skip-permissions-unsafe",
+          "model": "claude-sonnet-4-5-20250929"
+        }
+      },
+      "CLAUDE_HAIKU_4_5": {
+        "DROID": {
+          "autonomy": "skip-permissions-unsafe",
+          "model": "claude-haiku-4-5-20251001"
+        }
+      },
+      "CLAUDE_OPUS_4_1": {
+        "DROID": {
+          "autonomy": "skip-permissions-unsafe",
+          "model": "claude-opus-4-1-20250805"
+        }
+      },
+      "GLM_4_6": {
+        "DROID": {
+          "autonomy": "skip-permissions-unsafe",
+          "model": "glm-4.6"
+        }
+      }
     }
   }
-}
+}

crates/executors/src/executors/codex/normalize_logs.rs

@@ -41,7 +41,10 @@ use crate::{
         NormalizedEntryError, NormalizedEntryType, TodoItem, ToolResult, ToolResultValueType,
         ToolStatus,
         stderr_processor::normalize_stderr_logs,
-        utils::{ConversationPatch, EntryIndexProvider},
+        utils::{
+            ConversationPatch, EntryIndexProvider,
+            patch::{add_normalized_entry, replace_normalized_entry, upsert_normalized_entry},
+        },
     },
 };
 
@@ -311,40 +314,6 @@ enum UpdateMode {
     Set,
 }
 
-fn upsert_normalized_entry(
-    msg_store: &Arc<MsgStore>,
-    index: usize,
-    normalized_entry: NormalizedEntry,
-    is_new: bool,
-) {
-    if is_new {
-        msg_store.push_patch(ConversationPatch::add_normalized_entry(
-            index,
-            normalized_entry,
-        ));
-    } else {
-        msg_store.push_patch(ConversationPatch::replace(index, normalized_entry));
-    }
-}
-
-fn add_normalized_entry(
-    msg_store: &Arc<MsgStore>,
-    index_provider: &EntryIndexProvider,
-    normalized_entry: NormalizedEntry,
-) -> usize {
-    let index = index_provider.next();
-    upsert_normalized_entry(msg_store, index, normalized_entry, true);
-    index
-}
-
-fn replace_normalized_entry(
-    msg_store: &Arc<MsgStore>,
-    index: usize,
-    normalized_entry: NormalizedEntry,
-) {
-    upsert_normalized_entry(msg_store, index, normalized_entry, false);
-}
-
 fn normalize_file_changes(
     worktree_path: &str,
     changes: &HashMap<PathBuf, CodexProtoFileChange>,

crates/executors/src/executors/droid.rs

@@ -0,0 +1,172 @@
+use std::{path::Path, process::Stdio, sync::Arc};
+
+use async_trait::async_trait;
+use command_group::AsyncCommandGroup;
+use schemars::JsonSchema;
+use serde::{Deserialize, Serialize};
+use strum_macros::AsRefStr;
+use tokio::{io::AsyncWriteExt, process::Command};
+use ts_rs::TS;
+use workspace_utils::msg_store::MsgStore;
+
+use crate::{
+    command::CommandParts,
+    executors::{AppendPrompt, ExecutorError, SpawnedChild, StandardCodingAgentExecutor},
+    logs::utils::EntryIndexProvider,
+};
+
+pub mod normalize_logs;
+pub mod session;
+
+use normalize_logs::normalize_logs;
+
+use self::session::fork_session;
+
+// Configuration types for Droid executor
+#[derive(Debug, Clone, Copy, Serialize, Deserialize, PartialEq, TS, JsonSchema)]
+#[serde(rename_all = "kebab-case")]
+pub enum Autonomy {
+    Normal,
+    Low,
+    Medium,
+    High,
+    SkipPermissionsUnsafe,
+}
+
+fn default_autonomy() -> Autonomy {
+    Autonomy::SkipPermissionsUnsafe
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, TS, JsonSchema, AsRefStr)]
+#[serde(rename_all = "lowercase")]
+#[strum(serialize_all = "lowercase")]
+#[ts(rename = "DroidReasoningEffort")]
+pub enum ReasoningEffortLevel {
+    None,
+    Dynamic,
+    Off,
+    Low,
+    Medium,
+    High,
+}
+
+/// Droid executor configuration
+#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, TS, JsonSchema)]
+pub struct Droid {
+    #[serde(default)]
+    pub append_prompt: AppendPrompt,
+
+    #[serde(default = "default_autonomy")]
+    #[schemars(
+        title = "Autonomy Level",
+        description = "Permission level for file and system operations"
+    )]
+    pub autonomy: Autonomy,
+
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    #[schemars(
+        title = "Model",
+        description = "Model to use (e.g., gpt-5-codex, claude-sonnet-4-5-20250929, gpt-5-2025-08-07, claude-opus-4-1-20250805, claude-haiku-4-5-20251001, glm-4.6)"
+    )]
+    pub model: Option<String>,
+
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    #[schemars(
+        title = "Reasoning Effort",
+        description = "Reasoning effort level: none, dynamic, off, low, medium, high"
+    )]
+    pub reasoning_effort: Option<ReasoningEffortLevel>,
+
+    #[serde(flatten)]
+    pub cmd: crate::command::CmdOverrides,
+}
+
+impl Droid {
+    pub fn build_command_builder(&self) -> crate::command::CommandBuilder {
+        use crate::command::{CommandBuilder, apply_overrides};
+        let mut builder =
+            CommandBuilder::new("droid exec").params(["--output-format", "stream-json"]);
+        builder = match &self.autonomy {
+            Autonomy::Normal => builder,
+            Autonomy::Low => builder.extend_params(["--auto", "low"]),
+            Autonomy::Medium => builder.extend_params(["--auto", "medium"]),
+            Autonomy::High => builder.extend_params(["--auto", "high"]),
+            Autonomy::SkipPermissionsUnsafe => builder.extend_params(["--skip-permissions-unsafe"]),
+        };
+        if let Some(model) = &self.model {
+            builder = builder.extend_params(["--model", model.as_str()]);
+        }
+        if let Some(effort) = &self.reasoning_effort {
+            builder = builder.extend_params(["--reasoning-effort", effort.as_ref()]);
+        }
+
+        apply_overrides(builder, &self.cmd)
+    }
+}
+
+async fn spawn(
+    command_parts: CommandParts,
+    prompt: &String,
+    current_dir: &Path,
+) -> Result<SpawnedChild, ExecutorError> {
+    let (program_path, args) = command_parts.into_resolved().await?;
+
+    let mut command = Command::new(program_path);
+    command
+        .kill_on_drop(true)
+        .stdin(Stdio::piped())
+        .stdout(Stdio::piped())
+        .stderr(Stdio::piped())
+        .current_dir(current_dir)
+        .args(args);
+
+    let mut child = command.group_spawn()?;
+
+    if let Some(mut stdin) = child.inner().stdin.take() {
+        stdin.write_all(prompt.as_bytes()).await?;
+        stdin.shutdown().await?;
+    }
+
+    Ok(child.into())
+}
+
+#[async_trait]
+impl StandardCodingAgentExecutor for Droid {
+    async fn spawn(&self, current_dir: &Path, prompt: &str) -> Result<SpawnedChild, ExecutorError> {
+        let droid_command = self.build_command_builder().build_initial()?;
+        let combined_prompt = self.append_prompt.combine_prompt(prompt);
+
+        spawn(droid_command, &combined_prompt, current_dir).await
+    }
+
+    async fn spawn_follow_up(
+        &self,
+        current_dir: &Path,
+        prompt: &str,
+        session_id: &str,
+    ) -> Result<SpawnedChild, ExecutorError> {
+        let forked_session_id = fork_session(session_id).map_err(|e| {
+            ExecutorError::FollowUpNotSupported(format!(
+                "Failed to fork Droid session {session_id}: {e}"
+            ))
+        })?;
+        let continue_cmd = self
+            .build_command_builder()
+            .build_follow_up(&["--session-id".to_string(), forked_session_id.clone()])?;
+        let combined_prompt = self.append_prompt.combine_prompt(prompt);
+
+        spawn(continue_cmd, &combined_prompt, current_dir).await
+    }
+
+    fn normalize_logs(&self, msg_store: Arc<MsgStore>, current_dir: &Path) {
+        normalize_logs(
+            msg_store.clone(),
+            current_dir,
+            EntryIndexProvider::start_from(&msg_store),
+        );
+    }
+
+    fn default_mcp_config_path(&self) -> Option<std::path::PathBuf> {
+        dirs::home_dir().map(|home| home.join(".factory").join("mcp.json"))
+    }
+}

crates/executors/src/executors/droid/session.rs

@@ -0,0 +1,95 @@
+use std::{
+    fs, io,
+    path::{Path, PathBuf},
+};
+
+use serde_json::Value;
+use uuid::Uuid;
+
+pub fn fork_session(session_id: &str) -> io::Result<String> {
+    let root = sessions_root()?;
+    let source = find_session_file(&root, &format!("{session_id}.jsonl"))?;
+    let contents = fs::read_to_string(&source)?;
+    let ends_with_newline = contents.ends_with('\n');
+
+    let new_session_id = Uuid::new_v4().to_string();
+    let replaced = contents
+        .lines()
+        .enumerate()
+        .map(|(idx, line)| {
+            if idx == 0 {
+                replace_session_id(line, &new_session_id)
+            } else {
+                line.to_string()
+            }
+        })
+        .collect::<Vec<_>>()
+        .join("\n");
+
+    let mut output = replaced;
+    if ends_with_newline {
+        output.push('\n');
+    }
+
+    let destination = source
+        .parent()
+        .unwrap_or(root.as_path())
+        .join(format!("{new_session_id}.jsonl"));
+    fs::write(&destination, output)?;
+
+    if let Ok(settings_source) = find_session_file(&root, &format!("{session_id}.settings.json")) {
+        let settings_destination = settings_source
+            .parent()
+            .unwrap_or(root.as_path())
+            .join(format!("{new_session_id}.settings.json"));
+        let _ = fs::copy(settings_source, settings_destination);
+    }
+
+    Ok(new_session_id)
+}
+
+fn sessions_root() -> io::Result<PathBuf> {
+    dirs::home_dir()
+        .map(|home| home.join(".factory").join("sessions"))
+        .ok_or_else(|| io::Error::other("Unable to determine home directory"))
+}
+
+fn replace_session_id(line: &str, new_session_id: &str) -> String {
+    if let Ok(mut meta) = serde_json::from_str::<Value>(line)
+        && meta
+            .get("type")
+            .and_then(|value| value.as_str())
+            .map(|value| value == "session_start")
+            .unwrap_or(false)
+        && let Some(Value::String(id)) = meta.get_mut("id")
+    {
+        *id = new_session_id.to_string();
+        if let Ok(serialized) = serde_json::to_string(&meta) {
+            return serialized;
+        }
+    }
+
+    line.to_string()
+}
+
+fn find_session_file(root: &Path, filename: &str) -> io::Result<PathBuf> {
+    if root.join(filename).exists() {
+        return Ok(root.join(filename));
+    }
+
+    for entry in fs::read_dir(root)? {
+        let entry = entry?;
+        let path = entry.path();
+        if path.is_dir() {
+            let candidate = path.join(filename);
+            if candidate.exists() {
+                return Ok(candidate);
+            }
+        }
+    }
+
+    Err(io::Error::new(
+        io::ErrorKind::NotFound,
+        format!("Unable to locate {filename} in {}", root.display()),
+    ))
+}

crates/executors/src/executors/mod.rs

@@ -18,7 +18,7 @@ use crate::{
     command::CommandBuildError,
     executors::{
         amp::Amp, claude::ClaudeCode, codex::Codex, copilot::Copilot, cursor::CursorAgent,
-        gemini::Gemini, opencode::Opencode, qwen::QwenCode,
+        droid::Droid, gemini::Gemini, opencode::Opencode, qwen::QwenCode,
     },
     mcp_config::McpConfig,
 };
@@ -29,6 +29,7 @@ pub mod claude;
 pub mod codex;
 pub mod copilot;
 pub mod cursor;
+pub mod droid;
 pub mod gemini;
 pub mod opencode;
 pub mod qwen;
@@ -95,6 +96,7 @@ pub enum CodingAgent {
     CursorAgent,
     QwenCode,
     Copilot,
+    Droid,
 }
 
 impl CodingAgent {
@@ -125,6 +127,14 @@ impl CodingAgent {
                 self.preconfigured_mcp(),
                 false,
             ),
+            Self::Droid(_) => McpConfig::new(
+                vec!["mcpServers".to_string()],
+                serde_json::json!({
+                    "mcpServers": {}
+                }),
+                self.preconfigured_mcp(),
+                false,
+            ),
             _ => McpConfig::new(
                 vec!["mcpServers".to_string()],
                 serde_json::json!({
@@ -142,11 +152,12 @@ impl CodingAgent {
 
     pub fn capabilities(&self) -> Vec<BaseAgentCapability> {
         match self {
-            Self::ClaudeCode(_) => vec![BaseAgentCapability::SessionFork],
-            Self::Amp(_) => vec![BaseAgentCapability::SessionFork],
-            Self::Codex(_) => vec![BaseAgentCapability::SessionFork],
-            Self::Gemini(_) => vec![BaseAgentCapability::SessionFork],
-            Self::QwenCode(_) => vec![BaseAgentCapability::SessionFork],
+            Self::ClaudeCode(_)
+            | Self::Amp(_)
+            | Self::Codex(_)
+            | Self::Gemini(_)
+            | Self::QwenCode(_)
+            | Self::Droid(_) => vec![BaseAgentCapability::SessionFork],
             Self::CursorAgent(_) => vec![BaseAgentCapability::SetupHelper],
             Self::Opencode(_) | Self::Copilot(_) => vec![],
         }

crates/executors/src/logs/mod.rs

@@ -23,6 +23,22 @@ pub struct ToolResult {
     pub value: serde_json::Value,
 }
 
+impl ToolResult {
+    pub fn markdown<S: Into<String>>(markdown: S) -> Self {
+        Self {
+            r#type: ToolResultValueType::Markdown,
+            value: serde_json::Value::String(markdown.into()),
+        }
+    }
+
+    pub fn json(value: serde_json::Value) -> Self {
+        Self {
+            r#type: ToolResultValueType::Json,
+            value,
+        }
+    }
+}
+
 #[derive(Debug, Clone, Serialize, Deserialize, TS)]
 #[serde(tag = "type", rename_all = "snake_case")]
 #[ts(export)]

crates/executors/src/logs/utils/patch.rs

@@ -1,10 +1,12 @@
+use std::sync::Arc;
+
 use json_patch::Patch;
 use serde::{Deserialize, Serialize};
 use serde_json::{from_value, json, to_value};
 use ts_rs::TS;
-use workspace_utils::diff::Diff;
+use workspace_utils::{diff::Diff, msg_store::MsgStore};
 
-use crate::logs::NormalizedEntry;
+use crate::logs::{NormalizedEntry, utils::EntryIndexProvider};
 
 #[derive(Deserialize, Serialize, Debug, Clone, PartialEq, Eq, TS)]
 #[serde(rename_all = "lowercase")]
@@ -139,3 +141,37 @@ pub fn extract_normalized_entry_from_patch(patch: &Patch) -> Option<(usize, Norm
             .map(|entry| (entry_index, entry))
     })
 }
+
+pub fn upsert_normalized_entry(
+    msg_store: &Arc<MsgStore>,
+    index: usize,
+    normalized_entry: NormalizedEntry,
+    is_new: bool,
+) {
+    if is_new {
+        msg_store.push_patch(ConversationPatch::add_normalized_entry(
+            index,
+            normalized_entry,
+        ));
+    } else {
+        msg_store.push_patch(ConversationPatch::replace(index, normalized_entry));
+    }
+}
+
+pub fn add_normalized_entry(
+    msg_store: &Arc<MsgStore>,
+    index_provider: &EntryIndexProvider,
+    normalized_entry: NormalizedEntry,
+) -> usize {
+    let index = index_provider.next();
+    upsert_normalized_entry(msg_store, index, normalized_entry, true);
+    index
+}
+
+pub fn replace_normalized_entry(
+    msg_store: &Arc<MsgStore>,
+    index: usize,
+    normalized_entry: NormalizedEntry,
+) {
+    upsert_normalized_entry(msg_store, index, normalized_entry, false);
+}

crates/executors/src/mcp_config.rs

@@ -293,7 +293,7 @@ impl CodingAgent {
         use Adapter::*;
 
         let adapter = match self {
-            CodingAgent::ClaudeCode(_) | CodingAgent::Amp(_) => Passthrough,
+            CodingAgent::ClaudeCode(_) | CodingAgent::Amp(_) | CodingAgent::Droid(_) => Passthrough,
             CodingAgent::QwenCode(_) | CodingAgent::Gemini(_) => Gemini,
             CodingAgent::CursorAgent(_) => Cursor,
             CodingAgent::Codex(_) => Codex,

crates/server/src/bin/generate_types.rs

@@ -126,6 +126,9 @@ fn generate_types_content() -> String {
         executors::executors::copilot::Copilot::decl(),
         executors::executors::opencode::Opencode::decl(),
         executors::executors::qwen::QwenCode::decl(),
+        executors::executors::droid::Droid::decl(),
+        executors::executors::droid::Autonomy::decl(),
+        executors::executors::droid::ReasoningEffortLevel::decl(),
         executors::executors::AppendPrompt::decl(),
         executors::actions::coding_agent_initial::CodingAgentInitialRequest::decl(),
         executors::actions::coding_agent_follow_up::CodingAgentFollowUpRequest::decl(),
@@ -237,6 +240,10 @@ fn generate_schemas() -> Result<HashMap<&'static str, String>, serde_json::Error
             "copilot",
             generate_json_schema::<executors::executors::copilot::Copilot>()?,
         ),
+        (
+            "droid",
+            generate_json_schema::<executors::executors::droid::Droid>()?,
+        ),
     ]);
     println!(
         "✅ JSON schemas generated. {} schemas created.",

docs/agents/droid.mdx

@@ -0,0 +1,71 @@
+---
+title: "Factory Droid"
+description: "Set up Factory AI's Droid coding agent"
+icon: https://www.vibekanban.com/images/logos/factory-ai-logo-light.png
+---
+
+<Steps>
+<Step title="Install Droid CLI">
+  <Tabs>
+  <Tab title="macOS / Linux">
+  ```bash
+  curl -fsSL https://app.factory.ai/cli | sh
+  ```
+  </Tab>
+
+  <Tab title="Windows">
+  ```powershell
+  irm https://app.factory.ai/cli/windows | iex
+  ```
+  </Tab>
+  </Tabs>
+
+  For the latest installation instructions and platform-specific guidance, visit the [Factory AI CLI documentation](https://factory.ai/product/cli).
+</Step>
+
+<Step title="Authenticate with Factory AI">
+  To use Droid in Vibe Kanban, you must first authenticate:
+
+  1. Run `droid` to launch the interactive CLI
+  2. Use the `/login` command within Droid to authenticate
+
+  Alternatively, generate an API key from [Factory Settings](https://app.factory.ai/settings/api-keys) and set it as an environment variable:
+
+  ```bash
+  export FACTORY_API_KEY=fk-...
+  ```
+
+  For detailed authentication instructions, see the [Factory AI documentation](https://docs.factory.ai/factory-cli/getting-started/overview).
+</Step>
+
+<Step title="Start Vibe Kanban">
+  Once authenticated, launch Vibe Kanban:
+
+  ```bash
+  npx vibe-kanban
+  ```
+
+  You can now select Factory Droid when creating task attempts.
+</Step>
+</Steps>
+
+## Configuration Options
+
+Droid supports several configuration options in Vibe Kanban:
+
+- **Autonomy Level**: Controls permission level for file and system operations
+  - `normal`: Read-only mode – safe for reviewing planned changes without execution
+  - `low`: Basic file creation/editing while blocking system changes
+  - `medium`: Development operations with recoverable side effects
+  - `high`: Production operations with security implications or major side effects
+  - `skip-permissions-unsafe`: Bypasses all permission checks (use only in isolated environments) (default)
+
+- **Model**: Specify which model to use
+
+- **Reasoning Effort**: Control the reasoning depth
+  - `off`: Minimal reasoning
+  - `low`: Light reasoning
+  - `medium`: Balanced reasoning
+  - `high`: Deep reasoning for complex tasks
+
+These options can be configured when creating agent configurations in Vibe Kanban.

docs/configuration-customisation/agent-configurations.mdx

@@ -164,6 +164,22 @@ The profiles configuration uses a JSON structure with an `executors` object cont
 
   [View full documentation →](https://qwenlm.github.io/qwen-code-docs/en/cli/index)
 </Tab>
+
+<Tab title="DROID">
+  <ParamField path="autonomy" type="string">
+  Permission level: `"normal"`, `"low"`, `"medium"`, `"high"`, or `"skip-permissions-unsafe"`
+  </ParamField>
+
+  <ParamField path="model" type="string">
+  Specify which model to use
+  </ParamField>
+
+  <ParamField path="reasoning_effort" type="string">
+  Reasoning depth: `"off"`, `"low"`, `"medium"`, or `"high"`
+  </ParamField>
+
+  [View full documentation →](https://docs.factory.ai/factory-cli/getting-started/overview)
+</Tab>
 </Tabs>
 
 ### Universal Options

docs/docs.json

@@ -37,6 +37,7 @@
               "agents/amp",
               "agents/cursor-cli",
               "agents/opencode",
+              "agents/droid",
               "agents/ccr",
               "agents/qwen-code"
             ]

docs/integrations/vibe-kanban-mcp-server.mdx

@@ -88,6 +88,7 @@ When using `start_task_attempt`, the following executors are supported (case-ins
 - `cursor_agent` / `CURSOR_AGENT`
 - `qwen-code` / `QWEN_CODE`
 - `copilot` / `COPILOT`
+- `droid` / `DROID`
 
 ## Using the MCP Server
 

docs/supported-coding-agents.mdx

@@ -36,6 +36,10 @@ Cursor Agent CLI
 SST OpenCode
 </Card>
 
+<Card title="Droid CLI" icon="https://www.vibekanban.com/images/logos/factory-ai-logo-light.png" href="/agents/droid">
+Factory Droid
+</Card>
+
 <Card title="Claude Code Router" icon="https://www.vibekanban.com/images/logos/claude.svg#" href="/agents/ccr">
 Claude Code Router - orchestrate multiple models
 </Card>

frontend/src/components/ExecutorConfigForm.tsx

@@ -9,20 +9,11 @@ import { Card, CardContent } from '@/components/ui/card';
 import { Button } from '@/components/ui/button';
 import { Loader2 } from 'lucide-react';
 import { shadcnTheme } from './rjsf';
+import { BaseCodingAgent } from 'shared/types';
 // Using custom shadcn/ui widgets instead of @rjsf/shadcn theme
 
-type ExecutorType =
-  | 'AMP'
-  | 'CLAUDE_CODE'
-  | 'GEMINI'
-  | 'CODEX'
-  | 'CURSOR_AGENT'
-  | 'COPILOT'
-  | 'OPENCODE'
-  | 'QWEN_CODE';
-
 interface ExecutorConfigFormProps {
-  executor: ExecutorType;
+  executor: BaseCodingAgent;
   value: unknown;
   onSubmit?: (formData: unknown) => void;
   onChange?: (formData: unknown) => void;

shared/schemas/droid.json

@@ -0,0 +1,74 @@
+{
+  "$schema": "http://json-schema.org/draft-07/schema#",
+  "properties": {
+    "append_prompt": {
+      "title": "Append Prompt",
+      "description": "Extra text appended to the prompt",
+      "type": [
+        "string",
+        "null"
+      ],
+      "format": "textarea",
+      "default": null
+    },
+    "autonomy": {
+      "title": "Autonomy Level",
+      "description": "Permission level for file and system operations",
+      "type": "string",
+      "enum": [
+        "normal",
+        "low",
+        "medium",
+        "high",
+        "skip-permissions-unsafe"
+      ],
+      "default": "skip-permissions-unsafe"
+    },
+    "model": {
+      "title": "Model",
+      "description": "Model to use (e.g., gpt-5-codex, claude-sonnet-4-5-20250929, gpt-5-2025-08-07, claude-opus-4-1-20250805, claude-haiku-4-5-20251001, glm-4.6)",
+      "type": [
+        "string",
+        "null"
+      ]
+    },
+    "reasoning_effort": {
+      "title": "Reasoning Effort",
+      "description": "Reasoning effort level: none, dynamic, off, low, medium, high",
+      "type": [
+        "string",
+        "null"
+      ],
+      "enum": [
+        "none",
+        "dynamic",
+        "off",
+        "low",
+        "medium",
+        "high",
+        null
+      ]
+    },
+    "base_command_override": {
+      "title": "Base Command Override",
+      "description": "Override the base command with a custom command",
+      "type": [
+        "string",
+        "null"
+      ]
+    },
+    "additional_params": {
+      "title": "Additional Parameters",
+      "description": "Additional parameters to append to the base command",
+      "type": [
+        "array",
+        "null"
+      ],
+      "items": {
+        "type": "string"
+      }
+    }
+  },
+  "description": "Droid executor configuration",
+  "type": "object"
+}

shared/types.ts

@@ -34,9 +34,9 @@ export type ScriptRequest = { script: string, language: ScriptRequestLanguage, c
 
 export type ScriptRequestLanguage = "Bash";
 
-export enum BaseCodingAgent { CLAUDE_CODE = "CLAUDE_CODE", AMP = "AMP", GEMINI = "GEMINI", CODEX = "CODEX", OPENCODE = "OPENCODE", CURSOR_AGENT = "CURSOR_AGENT", QWEN_CODE = "QWEN_CODE", COPILOT = "COPILOT" }
+export enum BaseCodingAgent { CLAUDE_CODE = "CLAUDE_CODE", AMP = "AMP", GEMINI = "GEMINI", CODEX = "CODEX", OPENCODE = "OPENCODE", CURSOR_AGENT = "CURSOR_AGENT", QWEN_CODE = "QWEN_CODE", COPILOT = "COPILOT", DROID = "DROID" }
 
-export type CodingAgent = { "CLAUDE_CODE": ClaudeCode } | { "AMP": Amp } | { "GEMINI": Gemini } | { "CODEX": Codex } | { "OPENCODE": Opencode } | { "CURSOR_AGENT": CursorAgent } | { "QWEN_CODE": QwenCode } | { "COPILOT": Copilot };
+export type CodingAgent = { "CLAUDE_CODE": ClaudeCode } | { "AMP": Amp } | { "GEMINI": Gemini } | { "CODEX": Codex } | { "OPENCODE": Opencode } | { "CURSOR_AGENT": CursorAgent } | { "QWEN_CODE": QwenCode } | { "COPILOT": Copilot } | { "DROID": Droid };
 
 export type Tag = { id: string, tag_name: string, content: string, created_at: string, updated_at: string, };
 
@@ -230,7 +230,7 @@ executor: BaseCodingAgent,
  */
 variant: string | null, };
 
-export type ExecutorConfig = { [key in string]?: { "CLAUDE_CODE": ClaudeCode } | { "AMP": Amp } | { "GEMINI": Gemini } | { "CODEX": Codex } | { "OPENCODE": Opencode } | { "CURSOR_AGENT": CursorAgent } | { "QWEN_CODE": QwenCode } | { "COPILOT": Copilot } };
+export type ExecutorConfig = { [key in string]?: { "CLAUDE_CODE": ClaudeCode } | { "AMP": Amp } | { "GEMINI": Gemini } | { "CODEX": Codex } | { "OPENCODE": Opencode } | { "CURSOR_AGENT": CursorAgent } | { "QWEN_CODE": QwenCode } | { "COPILOT": Copilot } | { "DROID": Droid } };
 
 export type ExecutorConfigs = { executors: { [key in BaseCodingAgent]?: ExecutorConfig }, };
 
@@ -264,6 +264,12 @@ export type Opencode = { append_prompt: AppendPrompt, model?: string | null, age
 
 export type QwenCode = { append_prompt: AppendPrompt, yolo?: boolean | null, base_command_override?: string | null, additional_params?: Array<string> | null, };
 
+export type Droid = { append_prompt: AppendPrompt, autonomy: Autonomy, model?: string | null, reasoning_effort?: DroidReasoningEffort | null, base_command_override?: string | null, additional_params?: Array<string> | null, };
+
+export type Autonomy = "normal" | "low" | "medium" | "high" | "skip-permissions-unsafe";
+
+export type DroidReasoningEffort = "none" | "dynamic" | "off" | "low" | "medium" | "high";
+
 export type AppendPrompt = string | null;
 
 export type CodingAgentInitialRequest = { prompt: string,