Composable prompts for AI coding agents
Common issues and solutions.
Symptom: lfq shows nothing or errors.
Check if installed:
launchctl list | grep lfd
Reinstall:
lfd install
Run in foreground to debug:
lfd serve
Symptom: Task starts but never completes.
The task may be waiting for input. Use interactive mode:
lf review -i # interactive mode
Check if the coding agent is stuck on a permission prompt or clarifying question.
Symptom: Tasks fail with rate limit errors.
Claude, Codex, Gemini, and OpenCode have usage limits. Options:
lf review -m codexSymptom: Git worktree commands fail or show stale data.
List all worktrees:
git worktree list
Clean up stale entries:
git worktree prune
Remove merged worktrees:
lf ops wt prune
Symptom: lfq list shows a wave in WAITING state.
The loop hit its PR limit. Options:
Symptom: Task fails with context/token limit errors.
Reduce context:
lf review --no-lfdocs # skip repo docs
lf review --no-diff-files # skip branch files
lf review --area src/small/ # scope to specific area
Use summaries for large codebases:
lf ops summarize src/
See Configuration for context options.
Symptom: lf fails with “claude not found” or similar.
Run the setup wizard:
lf init
Check installation:
lf ops doctor