joenandez/caspar/caspar-fix

fix

Diagnose a bug to its root cause, get approval on the fix approach, then implement it test-first. Treat the current command $ARGUMENTS as the bug report.

Inputs

• bug_report ($ARGUMENTS): error/stack trace, repro steps, context. If empty, ask the user for error message, repro steps, and relevant context before proceeding.

Working Set

• The affected code paths and recent changes around the reported symptom (read just-in-time; do not inline file lists here).
• @caspar:analyst for parallel hypothesis investigation.

Outputs + DONE

A root-cause diagnosis and a verified fix. DONE when:

• Root cause is identified (not just the symptom suppressed) — name the cause, the affected files, and the fix approach.
• The fix was approved by the user before any code was written.
• The fix is implemented test-first (RED before GREEN — confirm the test fails for the right reason, then make it pass).
• Fix includes [🪳 TEMP {TOPIC}]-prefixed debug logging to confirm the fix and gather data if it fails.
• A closing summary lists what was delivered and the exact steps the user runs to validate.

Method / guardrails

1. Hypothesize. Brainstorm 5–7 possible sources; distill to the 1–2 most likely root causes, weighing the component's primary data flow, recent changes, error patterns, stack traces, and affected paths.
2. Investigate in parallel. Dispatch one @caspar:analyst per top hypothesis; each traces its hypothesis through the codebase (relevant code, recent commits, similar patterns, edge cases) and returns a compressed (~1–2K) finding in-thread — no scratch files.
3. Synthesize & gate. Present the root cause, affected files, and proposed approach. YOU MUST hold for user approval here — do NOT write code until approved (HoldForApproval).
4. Implement test-first. Per Skill(caspar-tdd): write the failing test, confirm RED, then implement the fix plus the [🪳 TEMP {TOPIC}] logging to GREEN. Address the root cause; do not paper over the symptom.

Handoff

Summarize what you delivered and the concrete validation steps. Then render the inline Next Steps line pointing to the natural follow-on (e.g. /caspar:test or /caspar:clean).

Escalate-If

• Bug report is empty or too thin to form hypotheses → ask the user for error, repro, and context.
• Investigation contradicts every hypothesis (no root cause found) → report findings and ask for more repro detail rather than guessing a fix.