agentleFS
nadavnavehnadavnaveh/skills/skills/objector/SKILL.md

objector

Skill2 starsChanged 24 days agoMITView on GitHub

Adds an objector to the agent's decisions. Before a decision that costs money or is hard to undo, a fresh subagent that did not make the decision argues that it is wrong, point by point. The first agent answers each point and changes the decision only for a fact it can check. Fixes "The agent made a call that cost money, and nobody argued with it." Use before a payment, a purchase, a booking, a price, a date or time someone will act on, a delete, a migration, a release, or any choice that is expensive to reverse. Skip it for cheap, reversible choices. Also on /objector.

The file

We split this skill into the rules it gives an agent

2 rules. About 0 are common, 2 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Only a fact you can check changes a decision.

    line 12
  2. Distinctive:

    Write the decision and its reasons in a few lines: what you decided, and each reason with the fact it rests on.

    line 17

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Objector

The agent made a call that cost money, and it was wrong. Nobody argued with it, because an agent checking its own decision agrees with itself. This skill adds an objector: a second agent whose only job is to prove the decision wrong before anyone acts on it.

The principles

  1. The agent that decided cannot be the one that objects.
  2. Only a fact you can check changes a decision. A good argument without one changes nothing.
  3. It costs tokens, so keep it for decisions that cost money or are hard to undo.

Steps

  1. Decide whether it is worth it. If the choice is cheap and easy to undo, skip this skill and say so in one line.
  2. Write the decision and its reasons in a few lines: what you decided, and each reason with the fact it rests on.
  3. Start a fresh objector that did not write the decision. In Claude Code, use a subagent (the Agent tool). Anywhere else, run a second claude -p with no shared context. Give it only the decision and the reasons. Its brief: "Make the strongest case that this decision is wrong, point by point. Tie each point to something checkable: a date, a number, a doc line or a command to run."
  4. Answer every point, one at a time. Check the fact it names yourself: open the doc, run the command or do the arithmetic. Mark each point as holds (the fact checks out and changes the decision), fails (the fact is wrong, and say what you found) or no fact (nothing to check, so nothing changes).
  5. Change the decision only for a point that holds, and name the fact that changed it.
  6. Show the person four things: the decision, the objections, your answer to each one, and what changed, or "nothing changed". Act only after they say yes.

Rules

  • The objector never sees the first agent's hidden reasoning. It sees only the decision and its reasons.
  • An objection without a checkable fact changes nothing.
  • This is a second opinion, not a proof. It does not make the decision safe.
  • Our run used Haiku on ten decisions with known answers, run three times. Alone, it got 29 of 30 right, and the one miss was the New York to Tel Aviv time after the US clock change. With a separate objector, the same question was missed once in a first answer, the objector caught it, and no right answer was flipped. It cost about 3x the tokens. One fix in 30 decisions is one event, not a rate.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.