Tab · Knowledge

Errors

The error-learning loop. Every mistake logged with a prevention rule, surfaced before you repeat it. Errors that recur become Lessons; Lessons regenerate into CLAUDE.md; the agent reads them on next run. The loop closes.

See it in motion

Where to find it

  • Localhost: /errors.html?repo=<id>
  • Source: .claude/brain/errors.md
  • Recurring: Same prevention rule fires ≥ 2× → flagged on the Today cockpit's Brain pulse panel.

What it does for you

Mistakes turn into rules.Each entry: what broke, what fixed it, what rule prevents it next time. The rule lands in `CLAUDE.md`; the next session walks in already knowing.
Recurring defects get attention.If the same rule fires twice in 30 days, the reflector promotes it to a recurring-rule fingerprint. Fix the rule, or replace it.
The hard-won knowledge compounds.Six months of errors.md is a brain that doesn't make beginner mistakes. The team brain (hosted rollup) sums across every bound device.

Configure

Nothing — errors.md is a flat .md file. New entries land via PR or the reflector.

Use it well

Add an error the moment you fix one. Be specific about the trigger and the fix; vague rules don't prevent recurrence. The reflector reads what you reject to bias future proposals — keep rejections loud.

Read more