| name | deploy |
| description | Deploy Codex or Gemini as working agents to implement tasks. Use when you want another AI to actually write code, not just review. (project) |
| allowed-tools | Bash, Read, Glob, Grep |
| user-invocable | true |
| proactive | true |
Deploy Agents
Deploy Codex or Gemini as working agents that can read, write, and modify files.
Agents
| Agent | Strength | Mode | Best For |
|---|---|---|---|
| Codex | Deep reasoning, careful implementation | codex exec --full-auto |
Complex refactors, nuanced logic |
| Gemini | Fast iteration, broad coverage | gemini --yolo |
File creation, straightforward tasks |
When to Use
- Parallelize work: Deploy both agents on independent subtasks
- Delegate implementation: Hand off a well-defined task while you work on something else
- Second implementation: Get a different approach to compare
- Bulk operations: Let an agent handle repetitive file changes
How to Run
When the user invokes /deploy, determine which agent(s) to use and what task to assign.
Step 1: Parse the Request
Identify:
- Task: What needs to be implemented
- Agent:
codex,gemini, orboth(default: ask user) - Scope: Which files/directories are relevant
Step 2: Gather Context
Read relevant files to include in the prompt. Keep context focused—agents work better with targeted information than entire codebases.
Key context files for SENTINEL:
C:\dev\SENTINEL\CLAUDE.md # Project overview
C:\dev\SENTINEL\AGENTS.md # Codex-specific guidance
C:\dev\SENTINEL\sentinel-agent\CLAUDE.md # Agent dev context
Step 3: Deploy Agent(s)
Deploy Codex
codex exec --full-auto -C "C:\dev\SENTINEL" "You are implementing a task for the SENTINEL project.
<context>
[Relevant code/docs - keep under 2000 lines]
</context>
<task>
[Clear, specific task description]
</task>
<constraints>
- Work within the existing project structure
- Follow patterns in CLAUDE.md and AGENTS.md
- Make minimal, focused changes
- Do not modify unrelated files
</constraints>
Implement this task. Create/modify files as needed."
Deploy Gemini
gemini --yolo "You are implementing a task for the SENTINEL project.
<context>
[Relevant code/docs - keep under 2000 lines]
</context>
<task>
[Clear, specific task description]
</task>
<constraints>
- Work within the existing project structure
- Follow patterns in CLAUDE.md
- Make minimal, focused changes
- Do not modify unrelated files
</constraints>
Implement this task. Create/modify files as needed."
Deploy Both (Parallel Tasks)
When tasks are independent, deploy both agents simultaneously on different subtasks:
# Terminal 1: Codex on subtask A
codex exec --full-auto -C "C:\dev\SENTINEL" "[Subtask A prompt]"
# Terminal 2: Gemini on subtask B
gemini --yolo "[Subtask B prompt]"
Step 4: Review Results
After agent(s) complete:
- Check
git statusfor changes made - Review the modified files
- Run tests if applicable:
cd sentinel-agent && pytest - Report what was done and any issues
Command Variations
Codex Options
| Flag | Effect |
|---|---|
--full-auto |
Sandbox + auto-approve (recommended) |
-s workspace-write |
Allow file writes in workspace |
-s read-only |
Read-only exploration |
-m o3 |
Use o3 model for harder tasks |
Gemini Options
| Flag | Effect |
|---|---|
--yolo |
Auto-approve all actions |
--approval-mode auto_edit |
Auto-approve edits only |
-s |
Run in sandbox |
-m gemini-2.5-pro |
Specify model |
Example Usage
Single Agent
User: /deploy codex Add a /history command to the CLI that shows recent session events
Then:
- Read
sentinel-agent/src/interface/cli.pyandcommands.py - Deploy Codex with focused context
- Review changes, run tests
Parallel Deployment
User: /deploy both Create unit tests for the state manager AND add docstrings to schema.py
Then:
- Assign Codex: unit tests for state manager
- Assign Gemini: docstrings for schema.py
- Deploy both simultaneously
- Merge results, resolve any conflicts
Quick Task
User: /deploy gemini Rename all instances of get_faction to fetch_faction
Then:
- Search for
get_factionoccurrences - Deploy Gemini with file list
- Verify renames are complete
Safety Notes
- Both agents run with file write access—review changes before committing
- Use
git diffto inspect all modifications - Agents may interpret tasks differently than expected—be specific
- For destructive operations, consider
--sandbox read-onlyfirst to see the plan
Differences from /council
/council |
/deploy |
|---|---|
| Gets opinions | Does work |
| Read-only | Read-write |
| Both agents always | Choose agent(s) |
| Synthesis focus | Execution focus |