/lessons — Per-module learned-lessons header

Round 5b1 introduced a deterministic lessons-header block at the top of each .claude/learned-lessons/<module>-lessons.md file. The retrospective agent appends raw lessons; an idle scheduler recomputes the header deterministically (no LLM in the loop). This skill surfaces that header without scrolling through long lessons files.

Usage

/lessons                              # all modules — one-line header per module
/lessons --module <name>              # full header for one module
/lessons --detail <name>              # header + last 10 raw lessons entries

What it shows

Per-module computed header:

• total_tasks — count of completed tasks attributed to the module.
• fail_rate_30d — weighted failure rate over the last 30 days, using a 0.95^days_ago exponential decay so recent failures dominate.
• recurring_patterns[] — array of {id, severity, occurrences}; patterns that recur across ≥3 tasks earn a stable id and surface here.
• recommended_tier_floor — minimum deviation tier (1–3) the developer should declare for new tasks in this module, derived from recurring pattern severity.
• recommend_pre_decompose — boolean; when true, the planner pre-flight gate splits tasks more aggressively before dispatch.

Use case

The planner reads recommended_tier_floor and recommend_pre_decompose during pre-flight. /lessons is the operator-side window into the same data — useful before manually running /start on a module known to have high fail rate, so you can size scope realistically.

Notes

The header is recomputed deterministically by the idle scheduler; values are reproducible from the same input lessons file. The retrospective agent never edits the header directly — it only appends raw lessons.