# AgentDelta

> AgentDelta — sandeep-alluru-agentdelta. Use this tool when you need to identify and diagnose discrepancies in AI agent behavior, comparing two agent runs to pinpoint where reasoning diverged, and detect prompt regression issues. It takes two agent run traces as input and outputs a detailed diff highlighting the exact points of divergence. Ideal for debugging and refining AI models, AgentDelta helps resolve inconsistencies and improves overall agent performance.

Canonical page: https://skillsregistry.net/skills/sandeep-alluru-agentdelta  
JSON: https://api.skillsregistry.net/v1/skills/sandeep-alluru-agentdelta

## Description

Agent trace diff and prompt regression detection — compares two agent runs to find exactly where reasoning diverged.

## Trust

- **Trust score (0–1):** 0.68
- **Verification tier:** scanned
- **Last scanned:** 2026-08-30

## Facts

- **Version:** 1.0.0
- **Skill type:** atomic
- **Execution layer:** mcp-remote
- **Runtime environment:** api
- **Category:** ai-ml
- **Updated:** 2026-08-30

## Source

- **Source listing:** [Glama](https://glama.ai/mcp/servers/gdaj7podo3)
- **Repository:** <https://github.com/sandeep-alluru/agentdelta>

## Use it

Resolve this record through the SkillsRegistry MCP server (no auth, read-only):

```
claude mcp add --transport http --scope user skillsregistry https://api.skillsregistry.net/mcp
```

```json
{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "get_skill",
    "arguments": {
      "slug": "sandeep-alluru-agentdelta"
    }
  }
}
```

REST: `GET https://api.skillsregistry.net/v1/skills/sandeep-alluru-agentdelta` · pull for local use: `GET https://api.skillsregistry.net/v1/skills/sandeep-alluru-agentdelta/pull`

---
SkillsRegistry indexes agent skills from public registries and GitHub. Skills we have analysed are scanned with Circle-IR and scored on six dimensions; each listing states its scan coverage. More: https://skillsregistry.net/llms.txt