agentleFS
KbediakoKbediako/evergreen-codex-skills/skills/challenge-assumptions/SKILL.md

challenge-assumptions

Skill5 starsChanged 2 months agoMITView on GitHub

Investigate uncertain or disputed causes, failed repairs, and high-impact diagnostic decisions. Skip straightforward failures already explained by authoritative evidence.

1 other project uses a version of this.

Adopted by

Most recently taken up by

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:

    Test the explanation that would change the next action.

    line 8
  2. Distinctive:

    Read only the relevant guidance:

    line 14

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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

Challenge assumptions

Test the explanation that would change the next action. Separate observations from causal inference, and use the smallest safe, authorized probe that can distinguish plausible causes.

Choose the depth

For a low-risk claim one probe can decide, keep the claim, disproof, and resulting decision together. Expand only an unresolved branch. For ambiguous, persistent, or high-impact failures, keep a compact hypothesis ledger with supporting evidence, contradictions, and a discriminating check. Reuse settled evidence while its scope, active system, and failure signature remain valid.

Read only the relevant guidance:

  • Evidence and probes for conflicting evidence, policy-shaped errors, or isolation across system boundaries.
  • Repair validation before state-changing diagnostics, containment, or repair, and after a failed or inconclusive repair.
  • Pushback and reporting when new user observations challenge a premise or the investigation needs a durable record.

Investigate

Identify the authoritative source and scope of each material observation. Check freshness, version, and the active consumer when they could change the conclusion. Label proxies and partial evidence. Activity counts establish progress only when they measure the requested outcome. A diagnosis inside an error message remains a claim; repeated views of one source are not independent evidence.

Keep only materially plausible alternatives. For each proposed probe, identify which possible result would change the next action or confidence. Its decision value must justify its cost, delay, and disruption. Do not run a fixed quota of probes or force alternatives after the cause is established.

A simpler path succeeding does not locate the cause. It weakens only explanations that predicted that path would fail. Identical error text or a shared mechanism does not establish a shared cause. Missing events count against a hypothesis only when the observation could reliably detect them.

When evidence contradicts a premise, revise that premise before repeating the conclusion. A bundled change may demonstrate recovery without identifying the cause. Before repeating or broadening a failed repair, establish whether the intended change reached the failing path and whether the behavior check validly tested it.

Complete the requested work

Stop diagnosis when the cause and next action are sufficiently supported, or no worthwhile safe probe can change the decision. If repair is already in scope, carry out the supported action and verify its effect. Do not turn an internal diagnostic checkpoint into a new approval gate. A diagnosis-only request does not authorize remediation.

Preserve evidence before a change can erase it. Urgent authorized containment may come first; record the observability it disturbs. Never bypass policy, authentication, or consent to test a theory, and do not recommend elevation or a broad remedy without evidence that narrower options are insufficient.

Claim repair completion only after the original failure path or a demonstrably scope-equivalent surrogate passes, effective active state matches the change, and affected regression boundaries have been checked proportionately. Otherwise report the verified result, leading inference, remaining uncertainty, and concrete next action or blocker.

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.