prompt-debugger

prompt-debugger is an agent for Claude Code from RadOrigin-LLC/RAD-Claude-Skills. It costs 456 tokens per session (3,319 once invoked), scanned A, original, Apache-2.0.

A diagnostic guide for finding why an AI prompt produced poor or unexpected output. It checks the prompt mechanically and links each observed failure to a specific cause and targeted fix.

In plain words
What is it for?
Use it to investigate prompts that an AI misunderstands, follows inconsistently, or answers incorrectly, with or without an example of the bad output.
Why use it?
It replaces vague prompt rewriting with a structured explanation of what went wrong and what small change may correct it.

Agent for Claude Code

Written for Claude Code: ${CLAUDE_PLUGIN_ROOT} variable. Also seen: model in frontmatter; mentions Claude Code.

Runs only inside its plugin — its command needs a path that Claude Code sets for a plugin’s own hooks and for nothing else. Install the plugin, not this.

Part of the rad-context-prompter plugin — 3 skills, 1 agent shipped together

Good fit Use it to investigate prompts that an AI misunderstands, follows inconsistently, or answers incorrectly, with or without an example of the bad output.

Compare 6 agents from other repositories ↓
Install

Getting it into your agent

This one installs as part of its plugin. Adding the marketplace and installing the plugin brings it with everything else the plugin ships.

Claude Code
/plugin marketplace add RadOrigin-LLC/RAD-Claude-Skills
Claude Code
/plugin install rad-context-prompter

Made for: Claude Code.

Or install rad-context-prompter, the plugin that ships this one along with the rest of its 3 skills, 1 agent.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for prompt-debugger

README.md
[![agentmods](https://agentmods.dev/badge/agents/radorigin-llc/rad-claude-skills/prompt-debugger.svg)](https://agentmods.dev/agents/radorigin-llc/rad-claude-skills/prompt-debugger)
Your own site
<a href="https://agentmods.dev/agents/radorigin-llc/rad-claude-skills/prompt-debugger"><img src="https://agentmods.dev/badge/agents/radorigin-llc/rad-claude-skills/prompt-debugger.svg" alt="Measured on agentmods" height="20"></a>
Per session 456 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 3,319 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00456 $0.03319
Opus 5 $0.00228 $0.01659
Sonnet 5 $0.00091 $0.00664
Haiku 4.5 $0.00046 $0.00332

Measured 8d ago against content hash 5c774c012557, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-08, from the pricing page.

Security

Grade A, and why

prompt-debugger scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 8d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

archive/plugins/rad-context-prompter/agents/prompt-debugger.md · 289 lines

How it starts

The opening of the file, as written. The whole thing — 289 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are a prompt failure analyst. When given a prompt (and optionally its bad output), you systematically diagnose why it failed and produce targeted, minimal fixes — not rewrites.

Your goal is precision: identify the specific mechanism that caused the failure, not vague observations about prompt quality. Every diagnosis must trace from symptom → root cause → fix.


Diagnostic Framework

Phase 0: Mechanical Pre-Pass

Before LLM judgment, run the bundled validators on the prompt text (save it to a temp file or pipe via stdin). Their findings seed Phase 1 — don't re-derive what they catch:

PY=$(command -v python3 || command -v python)
"$PY" "${CLAUDE_PLUGIN_ROOT}/scripts/lint-prompt.py" <prompt-file> --json
# Additionally, when the prompt is a loop prompt or goal/completion condition:
"$PY" "${CLAUDE_PLUGIN_ROOT}/scripts/check-goal.py" <prompt-file> --json

The scripts exit 1 when findings exist — expected, not an error. If Python is unavailable, proceed with LLM-only analysis.

Phase 1: Intake Classification

Classify the failure into exactly one primary category and up to two secondary categories:

Failure Taxonomy (8 categories, 44 specific patterns):

F1 — Output Shape Failures (the response has wrong format, length, or structure)

  • F1.1: No output format specified — model chose its own
  • F1.2: Format specified but not locked — model drifted mid-response
  • F1.3: Conflicting format signals — prompt style contradicts format instruction
  • F1.4: Implicit length — "summarize" without word/sentence count
  • F1.5: Wrong template for platform — prose prompt in Midjourney, flat prompt in Claude

F2 — Instruction Adherence Failures (model ignores or misinterprets instructions)

  • F2.1: Negative-only instructions — "don't do X" without stating what TO do
  • F2.2: Buried critical instruction — key constraint is after long context (attention decay)
  • F2.3: Contradictory instructions — two rules that can't both be followed
  • F2.4: Ambiguous scope — instruction can be interpreted multiple ways
  • F2.5: Over-prompting fatigue — too many MUST/NEVER/CRITICAL markers dilute all of them
  • F2.6: Model-specific instruction mismatch — CoT on reasoning models, verbose scaffolding on o3

Read the full file on GitHub · 289 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 8d ago First seen · 289 lines · 456 tokens per session scan A 5c774c012557

Subscribe to this mod's changes

prompt-debugger is an agent published in the GitHub repository RadOrigin-LLC/RAD-Claude-Skills (5 stars, last pushed 22d ago), licensed Apache-2.0. It adds 456 tokens to every session and 3,319 once invoked, about $0.0023 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.

Related

Other agents, from other repositories

prompts-guide

Interactive guide for using prompt-factory skill to generate mega-prompts. Helps choose from 69 presets or create custom prompts, select formats (XML/Claude/ChatGPT/Gemini), and explains usage. Use when user wants to generate production-ready prompts for any LLM.

alirezarezvani/claude-code-skill-factory · 60 tokens

rlm-factory-init-agent

Guided setup wizard for the rlm-factory plugin. Works standalone (O(1) keyword search across dense file summaries, zero external deps) or as Phase 1 of a Super-RAG stack with vector-db and/or obsidian-wiki-engine. Starts with a setup mode question so the user gets exactly what they need. Creates rlmprofiles.json and…

richfrem/agent-plugins-skills · 247 tokens

vector-db-init-agent

Guided setup wizard for the vector-db plugin. Works standalone (semantic search with zero external dependencies) or as part of a Super-RAG stack with rlm-factory and/or obsidian-wiki-engine. Starts with a setup mode question so the user gets exactly what they need. Installs Python dependencies, creates…

richfrem/agent-plugins-skills · 179 tokens

prompt-engineer

Use when: creating new prompts, optimizing existing prompts, reviewing prompt quality, designing agents or skills. Do NOT use for: code implementation (use domain expert), non-prompt tasks.

fusengine/agents · 41 tokens

super-rag-setup-agent

Multi-plugin initialization orchestrator for the Super-RAG knowledge stack. Provisions any combination of rlm-factory (RLM summary ledger), vector-db (semantic embedding index), and obsidian-wiki-engine (Karpathy-style LLM wiki) in a single guided session. Starts with a plugin audit and lets the user choose which…

richfrem/agent-plugins-skills · 271 tokens

semantic-drift-auditor

Modernization auditor designed to prevent business terminology decay and semantic drift. Compares migrated codebase entities, databases, and variables against a generated domain-lexicon.json mapped from specs/REQS.md. Trigger with "check semantic drift", "audit business vocabulary", or automatically before slice…

richfrem/agent-plugins-skills · 65 tokens