---
description: "Phase 1+2: Research the codebase and create a structured task plan for workers"
---

Analyze the following request and create an implementation plan:

$ARGUMENTS

## Process

### 0. Read host and persistence settings

```bash
python3 -c "
import json
with open('.multi-agent.json') as f:
    config = json.load(f)
persistence = config.get('persistence', {})
print(f'Host: {config.get("host", "claude-code")}')
print(f'Persist plan: {persistence.get("write_plan", False)}')
print(f'Persist tasks: {persistence.get("write_tasks", False)}')
print(f'Persistence dir: {persistence.get("dir", ".secure-coding")}')
"
```

### 1. Research

Explore the codebase before planning:
- use `Glob` to find relevant files
- use `Grep` to detect existing patterns, conventions, types, and interfaces
- use `Read` to understand similar implementations
- use the Agent tool with `subagent_type=\"Explore\"` if the repository is large
- if the active host is OpenCode or OmO, prefer host-native helpers such as `@explore` or `@general` for scoped read-only discovery when helpful

### 2. Clarifying questions

Before writing the plan, identify the ambiguous parts:
- What exact behavior is expected in edge cases?
- Are there performance, compatibility, or breaking-change constraints?
- Which files must not be modified?
- Are there new trust boundaries to analyze?

Present the questions and wait for answers before continuing.

### 3. Structured plan

Break the work into atomic tasks using this format:

```text
TASK 1: [descriptive title]
- Description: [exactly what to do]
- Files to modify: [complete list]
- Dependencies: none | Task N
- Can run in parallel with: Task N | none
- Context for the worker: [what the worker must know]
- Definition of done: [verifiable success criteria]

TASK 2: ...
```

If `.multi-agent.json` enables persistence, or the user explicitly asks to keep markdown artifacts, write:

- `.secure-coding/plan.md` with the approved plan
- `.secure-coding/tasks.md` with a checkbox task list derived from the plan

If the host is OmO and a native plan already exists in `.sisyphus/plans/`, do not overwrite it. Mirror the approved plan into `.secure-coding/` only when persistence is requested.

### 4. Initial security analysis

Identify these concerns before implementation:
- Are there new trust boundaries?
- Are there external inputs from users, APIs, or files?
- Does the change create new attack surfaces?
- Is `/threat-model` needed because of the architecture change?

### 5. Confirmation

Present the full plan and wait for approval before execution.
Suggest `/code` for implementation or `/full-cycle` for the complete workflow.