just-vibeGeneral · inspect by default
/debug
Investigate and explain a failure before changing code
Use to identify a cause and next experiment; fix applies a requested repair and repro builds a durable minimal reproduction.
Make it your own.
In Claude Code, use the slash command and add your context. In Codex, select debug from the just-vibe skill picker, then send the same brief.
Version 0.11.0 also supports /jv debug, /just-vibe debug and /jv:debug in Claude. See shortcut setup and context examples.
/just-vibe:debug Investigate intermittent checkout failures from these logs; do not change files./just-vibe:debug Diagnose a timeout that occurs only after a successful database write./just-vibe:debug Diagnose using redacted logs only; do not restart services or infer missing spans.What the agent does
- Inspect logs and code, then build a hypothesis list ranked by the observations that would discriminate between causes.
- Trace the first divergence from expected behavior with the smallest next experiment, using bounded probes rather than repeated full runs.
Inputs
- failure evidence, environment, and expected behavior.
Optional context: scope, references, constraints, successCriteria, environment, mode, budget.
Scope
- Reads
- Diagnosis and narrowing; no automatic code repair.
- Writes
- No product edits during diagnosis. Requested reproduction may create synthetic fixtures in owned temporary locations; tracked fixture-writing checks use an explicitly scoped apply verification. A request to repair selects the fix workflow.
- Mode
- Inspect; failure evidence, environment, and expected behavior. Local reproduction follows the bounded local execution rule; a reproduction that writes tracked files is a separately scoped apply run.
- Prerequisites
- Resolve the user brief and inspect the relevant project or supplied evidence. External capabilities are optional unless the selected action actually needs them.
Expected output
- Hypothesis table with supporting and contradicting evidence, the most supported cause, and the next targeted probe or proposed fix.
How the work is checked
- Eliminates an initially plausible false cause; insufficient evidence yields ranked hypotheses, not false certainty.
When to stop or clarify
- Avoid repeated identical probes. Do not mutate production or install debugging tools implicitly.
Handling missing context
- Infer
- Resolve the named files, existing scripts, current task and earlier corrections from the conversation and repository.
- Assume
- Use the narrowest interpretation that completes a reversible local task; state a consequential assumption once.
- Ask
- Ask when competing targets or incompatible success conditions would change the result; continue independent inspection first.
Technical guidance
- Evidence
- Collect logs, inputs, revision/environment and a bounded reproduction.
- Method
- Rank hypotheses by evidence and run the smallest experiment that distinguishes them before changing code.
- Pitfall
- Correlation with a recent edit or a noisy downstream stack trace does not establish causality.
- Check
- Explain the causal chain and the observation that rejected competing hypotheses; unresolved causes remain unresolved.
Situational decisions
When logs establish symptoms but not causation: Report competing hypotheses and the lowest-cost observation that separates them.
The coding agent follows this workflow using its available tools. Installation does not grant service access or guarantee an outcome. Read the compatibility notes.