Diagnose failures by testing rival hypotheses in parallel with read-only probes.
npx skills add https://github.com/m0n0x41d/haft --skill h-diagnose
Please help me install this Agent Skill. Skill: h-diagnose Source: https://github.com/m0n0x41d/haft/tree/main/internal/cli/skill/h-diagnose Command: npx skills add https://github.com/m0n0x41d/haft --skill h-diagnose
When a test, runtime, data, integration, or operational failure has multiple plausible causes, jumping to the first plausible story leads to wrong fixes. This Skill structures the diagnosis so the observable signal is stabilized first, rival explanations are tested against evidence, and inconclusive outcomes stay honest.
Ask the agent to diagnose the observed failure, for example: "Diagnose why the payment integration test fails intermittently in CI using h-diagnose."
High-intent search queries and answers about installing and using this skill.
Stabilize the observed signal first by stating the time window, scope, and reproducibility without naming a cause. Then generate at least three rival hypotheses that differ in kind, define a discriminating probe for each, and run read-only probes in parallel before ranking by evidence.
Rival-hypothesis testing is a diagnostic method where several distinct explanations for a failure are generated upfront and each is tested with probes designed to refute it. Hypotheses are ranked by evidence for, evidence against, and remaining uncertainty, keeping refuted rivals visible for fallback.
Use h-diagnose when a concrete failure has observable symptoms but an unclear cause across test, runtime, data, integration, or operational domains. Use h-frame instead for feature or redesign questions where no failure signal exists yet.
No, diagnosis does not authorize a fix. The Skill returns the leading explanation, concrete evidence, live rivals, and the next discriminating probe, and implementation proceeds only when the operator separately requests the change.
No, it defaults to a conversational, non-persistent diagnosis. A ProblemCard or hypothesis portfolio is persisted only on explicit save request or when a concrete receiving use requires the diagnosis to remain addressable, with provenance recorded.
thomasslobodaCorrelates logs, traces, metrics, and dashboards to identify root causes of incidents.
allexcdDiagnose, reproduce, and fix code defects from bug reports or error logs.
levanminhducIdentify and invert core assumptions to reveal hidden constraints and alternative approaches.
VKirillDiagnose root causes of bugs using a reproducible, testable debugging loop.
