Skip to content

SKILL.md

Latest commit

 

History

History
69 lines (49 loc) · 1.51 KB

SKILL.md

File metadata and controls

69 lines (49 loc) · 1.51 KB
name debug-performance
description Improve performance in Codex by measuring baseline behavior, isolating hot paths, implementing focused optimizations, and confirming wins with before/after numbers.
compatibility Codex CLI with terminal access, git, tokenlean CLI (npm i -g tokenlean)

Debug Performance (Codex)

Measure first. Optimize second.

Workflow

Measure -> Isolate -> Analyze -> Optimize -> Confirm

1. Measure

tl run "time <command>"   # Capture baseline with token-efficient output

Capture baseline metrics (latency, throughput, memory) before edits.

2. Isolate

tl parallel \
  "symbols=tl symbols <suspect-file>" \
  "complexity=tl complexity <suspect-file>" \
  "hotspots=tl hotspots"

Prioritize hot paths and repeated work.

3. Analyze

tl parallel \
  "snippet=tl snippet <function> <file>" \
  "flow=tl flow <function> <file>" \
  "deps=tl deps <file>"

Check for:

  • N^2 loops
  • Repeated expensive I/O
  • Heavy initialization on hot paths
  • Work that can be cached or deferred

4. Optimize

  • Use the smallest optimization with clear impact.
  • Preserve behavior and API.
  • Prefer algorithmic improvements over micro-optimizations.

5. Confirm

tl parallel "after=tl run 'time <same command>'" "tests=tl run 'npm test'"

Report before/after numbers. No numbers means no verified gain.

Tips

  • Do not optimize cold paths.
  • Keep benchmark input stable between runs.
  • If tradeoffs exist (speed vs memory), document them explicitly.