Debug Log Skill

debug-log-skill

[](LICENSE) [](editor-integrations/CLAUDE.md) [](editor-integrations/cursor/rules/debug-log.mdc) [](editor-integrations/AGENTS.md) [](github-actions/validate-debug-log.yml)

**Status:** v2.1 — the active knowledge graph plus a zero-dep `dls` CLI and a placeholder-sentinel guard that refuses TODO-shaped fake entries.

**A skill that turns every bug you fix into a rule the next person (or the next LLM session) won't have to rediscover.**

LLM coding assistants are fast, but they're forgetful. They'll ship the same subtle bug a second time, six weeks later, because the context from the first fix evaporated. This skill fixes that. Every bug you fix gets a short, structured entry in a file called `DEBUG_LOG.md` at your project root. Before writing new code, the agent greps that file to find rules that apply. Over a few weeks, the log becomes the highest-signal document in your repo — a living memory that humans and LLMs share.

The one-minute pitch

Three things compound when you use this:

1. You stop re-shipping the same bug. Every fix writes down what went wrong, what you believed that turned out to be false, and the rule that would have caught it. Next time anyone — you, a teammate, Claude, Cursor, Aider — touches that code, they find the rule before they re-break it.
2. Your agent stops hallucinating the same wrong API. LLM Hallucination or Assumption is a first-class bug category in this skill. When the agent invents a function that doesn't exist, you log it. The prevention rule ("this Compose modifier doesn't exist, use X") lives in the same