Skip to content
Back to skills

Escalate

ASecurity

Use when stuck in circles and want to hand off to another agent (Codex CLI) with full context

  • 666 stars
  • 0 votes
  • 0 copies
  • 3 views
  • Added February 8, 2026
ai-agentsgobashnode

Works with

  • cli

Security analysis

A100/100

Scanned February 12, 2026

npx -y skills add majiayu000/claude-skill-registry --skill escalate --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Escalate?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Escalate
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/majiayu000-escalate/badge)](https://www.skillsdirectory.com/skills/majiayu000-escalate)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: escalate
description: Use when stuck in circles and want to hand off to another agent (Codex CLI) with full context
disable-model-invocation: true
---

# Escalate

Generate a handoff prompt for another agent when you're stuck and need fresh perspective.

## When to Use

- You've tried multiple approaches and keep hitting the same walls
- You're going in circles and want a fresh perspective
- You need to context-switch but want to preserve progress for Codex CLI or another agent

## Workflow

```
/escalate → Choose detail level → Choose handoff mode → Analyze session → Generate prompt → pbcopy
```

## Step 1: Ask Detail Level

Use `AskUserQuestion` to ask:

```
How much context for the handoff?

1. Concise (~200-400 words) - Problem, key blockers, clear ask
2. Detailed (~500-800 words) - Above plus files, errors, code snippets
```

## Step 2: Ask Handoff Mode

Use `AskUserQuestion` to ask:

```
What should the next agent do?

1. Continue - Pick up where I left off and solve this
2. Fresh perspective - Ignore my approaches, try something new
3. Validate first - Confirm my understanding, then propose alternatives
4. Other - [let user specify custom instruction]
```

## Step 3: Analyze the Session

Review the conversation to extract:

| Section | What to Capture |
|---------|-----------------|
| **Problem** | The core objective - what you're trying to accomplish |
| **Approaches Tried** | Each attempt and why it didn't work |
| **Current Hypothesis** | What you think might be the root cause |
| **Relevant Files** | Key file paths that are central to the issue |
| **Error Messages** | Specific errors encountered (exact text) |
| **Constraints** | Things that must be preserved or avoided |

**For concise mode:** Focus on Problem, Approaches Tried, and one key blocker.

**For detailed mode:** Include all sections with code snippets and full error messages.

## Step 4: Generate the Prompt

### Template

```markdown
# Problem

[Clear statement of what you're trying to accomplish]

# Approaches Tried

- **Approach 1:** [what you tried]
  - Result: [what happened]
  - Why it failed: [analysis]

- **Approach 2:** [what you tried]
  - Result: [what happened]
  - Why it failed: [analysis]

[Continue for each significant approach]

# Current Hypothesis

[What you think might be the root cause or blocker]

# Relevant Context

**Files:**
- `path/to/file.ts` - [why it's relevant]
- `path/to/other.ts` - [why it's relevant]

**Errors:**
```
[Exact error messages]
```

**Constraints:**
- [Things that must be preserved]
- [Approaches to avoid and why]

# Your Task

[Based on handoff mode:]

- Continue: "Pick up where I left off. The context above shows what's been tried. Find a path forward and solve this."
- Fresh perspective: "Ignore the approaches above - they haven't worked. Look at this problem fresh and propose a different strategy."
- Validate first: "First, confirm whether my understanding of the problem is correct. Then propose an alternative approach."
- Other: [User's custom instruction]
```

## Step 5: Copy to Clipboard

```bash
echo "[generated prompt]" | pbcopy
```

Use a heredoc for multi-line content:

```bash
pbcopy << 'EOF'
[generated prompt content]
EOF
```

## Step 6: Confirm

Tell the user:

```
Handoff prompt copied to clipboard. Paste into Codex CLI to continue.
```

## Quality Checklist

- [ ] Problem statement is clear and self-contained
- [ ] Approaches include WHY they failed, not just WHAT was tried
- [ ] Hypothesis is specific, not vague
- [ ] File paths are absolute or repo-relative
- [ ] Error messages are exact, not paraphrased
- [ ] Constraints call out landmines to avoid
- [ ] Handoff instruction matches selected mode

## Common Mistakes

| Mistake | Fix |
|---------|-----|
| Vague problem statement | Be specific about the goal and success criteria |
| Listing approaches without failure analysis | Always explain WHY each approach didn't work |
| Missing context | Include enough that the next agent doesn't need to ask clarifying questions |
| Too much detail in concise mode | Ruthlessly cut to ~300 words for concise |
| Forgetting constraints | Warn about approaches that look promising but have been ruled out |

## Example Output (Concise)

```markdown
# Problem

Trying to get Jest tests to run in a WordPress plugin. Tests pass locally but fail in CI with "Cannot find module '@wordpress/scripts'" errors.

# Approaches Tried

- **Verified node_modules:** Confirmed package is installed, same versions local and CI
- **Cleared CI cache:** Rebuilt from scratch, same error
- **Checked path resolution:** Added moduleNameMapper in jest.config.js, no change

# Current Hypothesis

CI environment might have different NODE_PATH or module resolution behavior. The @wordpress/scripts package uses a nested dependency structure.

# Your Task

Pick up where I left off. The context above shows what's been tried. Find a path forward and solve this.
```

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.