Skip to main content

dev-trace

Evidence-driven causal tracing with competing hypotheses. Use when an observation has multiple plausible explanations and you need to systematically rank them with evidence FOR and AGAINST each. Pairs with @trail-tracer.

Datos de origen

Repositorio
evolution-foundation/evo-nexus
Última actividad en el origen
10 de abril de 2026 a las 16:24
Idioma detectado de SKILL.md
inglés
Estrellas
548
Forks
180

Opciones de instalación

De forma predeterminada está seleccionado el prompt que primero revisa el origen. Puedes cambiar a un comando directo o descargar una copia local.

Revisa los archivos de origen

Lee SKILL.md y los archivos complementarios que muestra SkillsMP antes de decidir si quieres instalarlo.

Mostrando SKILL.md

SKILL.md
Instrucciones de origen · Vista previa de solo lectura
name
dev-trace
description
Evidence-driven causal tracing with competing hypotheses. Use when an observation has multiple plausible explanations and you need to systematically rank them with evidence FOR and AGAINST each. Pairs with @trail-tracer.
# Dev Trace Derived from oh-my-claudecode (MIT, Yeachan Heo). Adapted for the EvoNexus Engineering Layer. Causal investigation with multi-hypothesis discipline. Generate at least 2 explanations, collect evidence FOR and AGAINST, rank by strength, end with a discriminating probe. ## Use When - Observation has multiple plausible causes - "Why did this happen?" — and you don't want to jump to the first answer - Multi-causal problem (intermittent bugs, regional anomalies, performance puzzles) - Need to preserve uncertainty rather than collapse to one answer prematurely ## Do Not Use When - Cause is obvious → just fix it via `@hawk-debugger` - Single-cause bug with clear stack trace → `@hawk-debugger` - Just needs codebase search → `@scout-explorer` ## Workflow Delegate to `@trail-tracer`: 1. **OBSERVE** — restate the observation precisely, no interpretation 2. **FRAME** — define the exact "why" question 3. **HYPOTHESIZE** — generate 2+ competing explanations 4. **GATHER EVIDENCE** — for each, collect evidence FOR and AGAINST in parallel 5. **APPLY LENSES** — systems / premortem / science 6. **REBUT** — strongest alternative challenges the leader 7. **RANK** — down-rank contradicted, weak-assumption, failed-prediction explanations 8. **SYNTHESIZE** — best explanation, explicitly provisional if needed 9. **PROBE** — name the critical unknown and the discriminating next experiment ## Output Saved to `workspace/development/debug/[C]trace-{topic}-{date}.md` using the `dev-trace-report.md` template. ## Pairs With - `@trail-tracer` (the executor) - `@hawk-debugger` (when trace converges to a clear bug) - `@apex-architect` (when trace reveals architectural issues)
Ver en GitHub