agentleFS
youcislayoucisla/Agent-Foundry/skills/core/verify-first/SKILL.md

verify-first

Skill1 starsChanged 23 days agoMITView on GitHub

'Every assertion about the codebase, product, or market is a hypothesis.

The file

We split this skill into the rules it gives an agent

9 rules. About 0 are common, 9 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈9

    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:

    Every assertion is a hypothesis.

    line 15
  2. Distinctive:

    Always active. This is the meta-discipline that governs all work.

    line 19
  3. Distinctive:

    Always. This is the foundation.

    line 22
  4. Distinctive:

    When the user says "the code does X"

    line 24
  5. Distinctive:

    When a plan depends on an assumption about the system

    line 25
  6. Distinctive:

    Before accepting anything as fact, mentally tag it:

    line 31
  7. Distinctive:

    When you discover a new type of stale claim, add it to known-patterns so future sessions don't fall for the same trap.

    line 56
  8. Distinctive:

    Verify the world, not the self-report: re-read the stored row, artifact, or response.

    line 82
  9. Distinctive:

    When two runs are compared, prove the confounders match: same config except the tested flag, same provider stub, no retries or preemption.

    line 86

≈ “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

Verify First

Every assertion is a hypothesis. Verify before committing.

When to Use

Always active. This is the meta-discipline that governs all work.

  • Always. This is the foundation.
  • Especially: when inheriting work from another agent or audit
  • When the user says "the code does X"
  • When a plan depends on an assumption about the system

Procedure

1 — Tag every input as a hypothesis

Before accepting anything as fact, mentally tag it:

  • Untested assumption (brief says X)
  • Verified fact (I checked: X is true)
  • Stale truth (was X, now Y)

2 — Triangle verification

For any important claim, verify from at least 2 independent sources:

Claim: "The Stripe account is ready for billing"
- Source 1: Stripe MCP (products list → 0 products)
- Source 2: Code search (VITE_BILLING unset, test key only)
- Conclusion: ❌ wrong — wrong account entirely

Best sources to triangulate:

  1. Live DB query (what's actually stored)
  2. Live API call (what the server actually does)
  3. Source code (what the repo says)
  4. Browser (what users actually see)
  5. Git log (when was it last changed)

3 — Log and generalize

When you discover a new type of stale claim, add it to known-patterns so future sessions don't fall for the same trap.

Composes With

  • re-verify-findings — for audit/bug claim context
  • measure-first — adds live data to the triangulation
  • read-before-build — applies the same discipline to plan claims

Anti-patterns

  • Repeating a brief's claim without checking it this session.
  • Trusting an agent's report that a file or row was written.
  • Calling a compile a behavior check.
  • Collapsing timeout, failure, and low score into one status.

Verification Checklist

  • Every important claim has a source and a verdict.
  • Artifacts were re-read, not trusted from a report.
  • Untouched files are byte-identical where claimed.
  • Failure states are distinguished exactly.
  • Unverifiable claims are reported as unverified.

Verify the Artifact

  • Verify the world, not the self-report: re-read the stored row, artifact, or response. An agent's claim that it wrote something is not evidence.
  • Assert untouched files are byte-identical, not merely present.
  • A guard only guards if the intended regression fails it. Introduce the regression once, watch the test go red, then revert.
  • Preserve exact failure status: timeout, unparseable, call failure, and low score are four different facts.
  • When two runs are compared, prove the confounders match: same config except the tested flag, same provider stub, no retries or preemption.

Evidence Requirement

A verification claim needs a reference: a file path with line range, a command output chunk, or a stored execution id. A prose assertion is not evidence. If the reference cannot be produced, report the claim as unverified.

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.