agentleFS
youcislayoucisla/Agent-Foundry/skills/core/pushback-when-wrong/SKILL.md

pushback-when-wrong

Skill1 starsChanged 23 days agoMITView on GitHub

When the brief contradicts verified reality, push back with evidence.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    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:

    Before building anything expensive

    line 27
  2. Distinctive:

    Before designing the plan, treat each claim as a hypothesis:

    line 33
  3. Distinctive:

    Every plan must end with two subsections:

    line 47

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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

Pushback When Wrong

Silent execution of wrong premises is the #1 cause of wasted engineering weeks.

When to Use

Activates on any planning, strategy, or audit task where the user has expressed firm opinions about what's needed.

  • User gives a brief with specific claims about what's wrong
  • User asks for feature X (you suspect Y matters more)
  • User says "do this in this order" (data suggests different order)
  • User assumes something about their own codebase that may be wrong
  • Before building anything expensive

Procedure

1 — Verify every claim in the brief

Before designing the plan, treat each claim as a hypothesis:

| Claim | Verification | Result |
|-------|--------------|--------|
| "45 migrations" | COUNT(supabase/migrations/*.sql) | Actual: 24 |
| "13 tables" | COUNT(tables in public) | Actual: 14 |
| "iOS missing" | ls ios/ | Exists |

Tag them: ✅ confirmed / 🔧 fixed / ❌ wrong / 🆕 new finding.

2 — Pushback section in the plan output

Every plan must end with two subsections:

What you're wrong about (with evidence):

  • Claim 1: Reality — Why it matters
  • Claim 2: Reality — Why it matters

What you haven't thought of (top 5-10):

  1. [Surprising gap]: [why it matters + what to do]

3 — Be specific, not diplomatic

Vague pushback ("this might be wrong") is noise. Specific pushback ("you said 45 migrations, I counted 24") is useful. Always include the original claim, the evidence (query result, file count, screenshot), and why the gap matters.

4 — Propose the alternative

Pushback without an alternative is just complaining. Always end with: "instead, do X because Y."

Tone

Direct, evidence-based, not aggressive. The user knows they're fallible; they hired you to check, not to agree.

Composes With

  • verify-first — provides the evidence base
  • measure-first — provides the data for "your order is wrong"
  • show-your-work — documents the pushback in the final report

Anti-patterns

  • Agreeing with a wrong premise to avoid friction.
  • Vague disagreement without a count, query, or screenshot.
  • Pushback with no proposed alternative.
  • Burying one valid correction in ten speculative ones.

Verification Checklist

  • Each challenged claim has evidence.
  • The response names what the user missed.
  • Each pushback ends with a concrete alternative.
  • Tone is direct and non-aggressive.
  • New findings are separated from corrections.

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.