agentleFS
Sign inStart free
maroofiumsmaroofiums/Evidence-First-Debugger/AGENTS.md

Evidence-First-Debugger

AGENTS.md0 starsChanged 16 days agoMITView on GitHub

Core rule: a fix is a hypothesis until it has been verified by execution. Reading code is not verification. Full reference material: .skills/evidence-first-debugger/

We split this AGENTS.md into the rules it gives an agent

4 rules. About 0 are common, 4 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    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:

    Reproduce: Run the failing command and capture real output.

    line 10
  2. Distinctive:

    Verification gate: Run the original reproduction, the new test, and the surrounding suite.

    line 16
  3. Distinctive:

    Do not delete, skip, or loosen a failing test to make it pass.

    line 29
  4. Distinctive:

    If you cannot run code, say so, still follow the procedure, and report UNVERIFIED with commands.

    line 30

≈ “Common” is an estimate: at least 1 in 5 of the 36,368 AGENTS.md 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
AGENTS.md

Evidence-First Debugging Protocol

Core rule: a fix is a hypothesis until it has been verified by execution. Reading code is not verification.

When this applies

Any bug, error, traceback, failing test, wrong output, flaky test, crash, or performance regression.

Procedure

  1. Symptom. Record expected vs actual, the exact error text (verbatim), environment, frequency, and recent changes.
  2. Reproduce. Run the failing command and capture real output. Reduce it to a minimal reproduction. Save the command.
  3. Evidence. Collect observations (stack trace, actual values, git diff, versions, logs) before naming any cause. Label each fact with its source.
  4. Hypotheses. List at least two to three candidates, ranked. For each: evidence for, evidence against, and a cheap falsifying check. Run the cheapest check first. Mark each confirmed, eliminated, or inconclusive.
  5. Root cause. Declare one only when a check confirms it, ideally by toggling the suspected factor.
  6. Minimal fix. Change only what the root cause requires. No drive-by refactors. Do not patch the symptom (for example a blanket try/except or None check) without explaining why the bad state arose.
  7. Regression test. It must fail on the original code and pass on the fixed code. It must be deterministic. Confirm both states by running them.
  8. Verification gate. Run the original reproduction, the new test, and the surrounding suite. For flaky bugs, repeat enough runs to be meaningful (about 3/p runs, where p is the original failure rate) and state the count.

Required final status (exactly one)

  • VERIFIED: reproduction and tests were executed after the fix and passed. Show the commands and output.
  • PARTIALLY VERIFIED: some checks ran; state exactly which did not.
  • UNVERIFIED: nothing was executed. Provide the exact commands the user should run and the output that distinguishes success from failure.

Forbidden unless status is VERIFIED

"This fixes it", "the bug is fixed", "this should resolve it", "that was the problem". Use calibrated language such as "the evidence points to" or "this is expected to fix it, pending the commands below".

Other rules

  • A changed error message is not a fix.
  • A single passing run does not verify a fix for an intermittent bug.
  • Do not delete, skip, or loosen a failing test to make it pass.
  • If you cannot run code, say so, still follow the procedure, and report UNVERIFIED with commands.
  • Compact report for small bugs: symptom, evidence, root cause, fix, test, status.

Full reference material: .skills/evidence-first-debugger/

Download now (ZIP)

  • Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.

    Of the AGENTS.md files we read, the 28,693 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.

  • In this repository3 other files it gives its agents.

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.

Check your setup, free