| name | systematic-debugging |
| description | Use a strict observe-hypothesize-trace-verify-fix loop when debugging; avoid random changes and context loss. Use when the user is debugging, investigating a failure, or fixing a bug. |
Systematic Debugging
Use this skill when debugging failing tests, runtime errors, or unexpected behavior. It keeps the agent from guessing, repeating failed attempts, or losing context across sessions.
When to use
- User says they're debugging, investigating a failure, or fixing a bug.
- A test or build is failing and the user wants help finding the cause.
- The task involves "extend", "core tests", or "all-less" and may span multiple packages.
Core phases (do not skip)
- Phase 1 (Root cause investigation) — Reproduce the problem, gather evidence, inspect recent diffs, and log what is actually happening vs what should happen.
- Phase 2 (Pattern analysis) — Find similar working code; compare differences, dependencies, and assumptions before touching code.
- Phase 3 (Hypothesis + test) — Form a single, testable hypothesis and verify it with the smallest possible change. If it fails, revise and repeat.
- Phase 4 (Implementation) — Fix only the confirmed root cause, create/keep a failing test, re-run the focused test, and then run the broader suite.
- Update state — Record progress (
.cursor/PROJECT_STATE.md current-focus block or active area plan), remove .only filters, and document next steps.
Anti-patterns (do not do)
- Random changes — Trying multiple unrelated edits hoping one works.
- No trace — Assuming which code path runs or what a variable is without logging or reading the code path.
- Long session without state update — Debugging for a long time without writing "what we tried" and "next step" into PROJECT_STATE (and the area’s active plan file, if any).
- Skipping build — Changing
packages/core and running jess tests without pnpm --filter @jesscss/core build first.
- Logging wrong things — Using
console.log (suppressed in Vitest), or JSON.stringify on AST/parser nodes (circular refs, frame state). Use syncLog() and only primitive values.
Project-specific
- Package scripts: Run tests/builds from the package directory or with
pnpm --filter @jesscss/<pkg>. See .cursor/rules/30-tests.mdc.
- State and commands: Read
.cursor/PROJECT_STATE.md at the start. Use
scoped rules and canonical package docs for area-specific context; avoid old
root-level plan files unless they are explicitly marked active. Use
/start-debugging to begin, /run-baseline to get a clean pass/fail report,
/update-debug-state at end of session.
- Isolation: Use
it.only() or describe.only() to focus one test; remove before commit. See .cursor/rules/30-tests.mdc.
References
- State and session discipline:
.cursor/PROJECT_STATE.md and /update-debug-state
