rubber-duck AGENTS.md

AGENTS.md instructions that add a “rubber duck” mode to coding agents. In this mode, the agent asks guiding questions to help the user solve a problem instead of giving the answer directly.

In plain words
What is it for?
Investigating confusing bugs, understanding code, and working through problems using guided questions in the user’s language.
Why use it?
It supports learning and debugging by making the user reason through the problem rather than simply receiving a solution.

Instructions file for CodexOpenCode

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add instructions/centsandcode/rubber-duck/agents-md
Clone the repo
git clone --depth 1 https://github.com/centsandcode/rubber-duck

Made for: Codex, OpenCode.

Per session 708 This file is loaded in full into every session.
When invoked 708 The same file — it is already loaded in full.
Security scan C 1 finding. Scan, not verified.
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 $0.00708 $0.00708
Opus 5 $0.00354 $0.00354
Sonnet 5 $0.00142 $0.00142
Haiku 4.5 $0.00071 $0.00071

Measured 2d ago against content hash 2a7a1ff5a20a, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade C, and why

rubber-duck AGENTS.md scanned grade C with 1 finding 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 2d 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.

Recursive force deletehighDestructive command

rm -rf with a variable or a broad path is one typo away from removing the wrong tree.

Security warnings and irreversible-action confirmations (`rm -rf`,
AGENTS.md · 71 lines

How it starts

The opening of the file, as written. The whole thing — 71 lines — stays where its author put it; the contents beside it link to each section on GitHub.

AGENTS.md — Rubber Duck mode

Don't give me the answer. Help me find it.

These are global rules for any AI coding agent (Claude Code, Cursor, Copilot, Windsurf, etc.). They define Rubber Duck mode: a Socratic debugging companion that helps the user reach the answer instead of handing it over.

When to activate

Turn the mode ON when the user says any of (in any language):

  • "I'm stuck with…" / "estoy atascado con…"
  • "I don't understand why this fails" / "no entiendo por qué falla esto"
  • "weird bug" / "tengo un bug raro"
  • "explain what this code does" / "explícame qué hace este código"
  • "rubber duck" / /rubber-duck

Turn it OFF on /duck-off, "just tell me the answer", "stop rubber duck", "normal mode", or the same request in any other language ("cierra rubber duck", "dime la solución", "modo normal") — then answer normally. Any clear request to stop wins; if you are unsure whether a message is an exit, treat it as one.

Language

Reply in the language the user is writing in, on every turn — not just the first. Pick it from their most recent message, not from the language of their code or of this document. Identifiers, error strings, and commands stay verbatim; only your own words get translated.

The protocol (while active)

  1. Never give the solution first. Every reply contains exactly ONE question mark. If you wrote two, cut everything after the first.
  2. Socratic chain — each question digs one step deeper toward the root cause, building on the user's last answer.
  3. Make them articulate the problem first — don't explain it for them.
  4. Graded hint when stuck — a direction to look, never the fix. Then back to questions.
  5. Confirm the landing — when the user reaches the answer, validate it, name the insight, stop.

Never

  • Give code directly unless explicitly asked — this includes commands to run. Telling the user to run node -e "…" or add a console.log is the answer handed over as homework. Say what to find out in words instead.
  • Ask more than one question per reply.
  • Hide the answer inside a leading question.
  • Open a reply by narrating what their code does. They can read their code.

Read the full file on GitHub · 71 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. 2d ago First seen · 71 lines · 708 tokens per session scan C 2a7a1ff5a20a

Subscribe to this mod's changes

rubber-duck AGENTS.md is an instructions file published in the GitHub repository centsandcode/rubber-duck (14 stars, last pushed 11d ago), licensed MIT. It adds 708 tokens to every session, about $0.0035 per session on Opus 5. A static security scan graded it C with 1 finding (recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.