Update to Skillquiver

{
  "description": "Finds root causes through reproduction and falsifiable experiments. Use when a failure, flaky behavior, or regression lacks a demonstrated cause.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 223
    },
    {
      "relative_path": "condition-based-waiting.md",
      "size_in_bytes": 3079
    },
    {
      "relative_path": "defense-in-depth.md",
      "size_in_bytes": 3640
    },
    {
      "relative_path": "find-polluter.sh",
      "size_in_bytes": 2859
    },
    {
      "relative_path": "root-cause-tracing.md",
      "size_in_bytes": 5281
    }
  ],
  "name": "diagnose-systematically",
  "skill_md_contents": "---\nname: diagnose-systematically\ndescription: Finds root causes through reproduction and falsifiable experiments. Use when a failure, flaky behavior, or regression lacks a demonstrated cause.\n---\n\n# Diagnose Systematically\n\nFind the cause before proposing a fix. Diagnosis does not authorize implementation.\n\n## Preserve evidence\n\nKeep an append-only investigation log with the symptom, commands, exit codes,\ndecisive output, hypotheses, experiments, and verdicts. Put it in a scratch\ndirectory unless the user requested a versioned artifact. If the user requests\nread-only work or forbids edits, report that trail through command output and\nthe final response; do not create any file, including a scratch or temporary log.\n\n## Evidence loop\n\n1. **State the contract.** Name the involved components, their expected\n   behavior, and the exact observed symptom. Verify environment assumptions\n   such as build, branch, configuration, and dependency versions.\n2. **Make it fail.** Run the smallest unattended signal that reaches the real\n   symptom: focused test, CLI or HTTP assertion, headless UI check, trace\n   replay, minimal harness, or seeded stress runner. A process starting, a\n   string existing, or a silently skipped test is not a reproduction.\n3. **Minimize.** Remove one input, dependency, configuration value, or step at\n   a time. Keep a removal only when the same symptom remains. For intermittent\n   defects, measure a reproduction rate instead of treating one pass as proof.\n4. **Form falsifiable hypotheses.** First compare a working example of the same\n   pattern with the failing path. Write only credible hypotheses, each with a\n   distinguishing prediction and falsifier.\n5. **Probe one variable.** Observe state at the earliest boundary where the\n   predictions diverge. Revert probes or instrumentation that did not explain\n   the symptom before trying the next one.\n6. **Claim only what discriminates.** A cause is verified only when its\n   prediction is observed through the symptom-specific signal and plausible\n   alternatives are ruled out. Separate verified cause, inference, and\n   proposed behavior change in the report.\n\nA useful signal is symptom-specific, red-capable, deterministic or measured,\nfast enough to repeat, and agent-runnable. If no such signal can be built,\nreport the attempts and request the smallest missing artifact or access.\n\n## Scale the investigation deliberately\n\nKeep one investigator for a small reproduced defect. Use parallel read-only\nlanes only when several components remain plausible, the regression window is\nuncertain, or intermittency supports independent probes. Give every lane the\nsame symptom packet and one seam: reproduction scope, first code-path\ndivergence, recent-change regression, or proof observability. Require a\nhypothesis, prediction, falsifier, evidence, missing evidence, smallest next\nprobe, and confidence. Merge duplicate theories and run the cheapest\ndiscriminating probe; ranking never proves a cause.\n\nUse these resources only when their condition matches:\n\n- Deep call-stack symptom: [root-cause-tracing.md](root-cause-tracing.md).\n- Arbitrary delays or flaky async tests:\n  [condition-based-waiting.md](condition-based-waiting.md).\n- Test-order pollution: run [find-polluter.sh](find-polluter.sh).\n- Proven source fix that needs layered guards:\n  [defense-in-depth.md](defense-in-depth.md).\n\n## Fix only when authorized\n\nWhen the user requested a fix, turn the minimized reproduction into a\nregression test at the highest public seam that exercises the defect. Record it\nfailing for the expected reason before changing production code. Implement the\nsmallest fix, rerun the identical test, then rerun the original unminimized\nsignal. Where cheap, revert the fix once to demonstrate the failure returns.\n\nDo not substitute a shallow test when no correct seam exists; state the\nlimitation. Remove temporary instrumentation and harnesses. After three failed\nfix attempts, stop and question the causal model or architecture instead of\nstacking a fourth guess.\n\n## Close\n\nReport the supported cause, discriminating evidence, exact checks and results,\nauthorized fix if any, removed instrumentation, and unverified surfaces. Do not\ncall an unexplained recovery a fix.\n"
}