lessonlisted

# lesson Skill ## Commands - `/lesson add <tag> <one-line description>` — manually capture a lesson right now - `/lesson note <one-line observation>` — park an UNRESOLVED observation; changes nothing - `/lesson-prune` — flag stale lessons for deletion, and triage parked notes ## The three-way distinction (never confuse these) | Command | Meaning | What changes | |---|---|---| | `/lesson note` | "I *saw* something — don't lose it" | Nothing. One line lands in an inbox. | | `/lesson add` | "I *learned* something — resolved symptom + fix" | Knowledge injected into future plans/agents via the bank. | | `/boiler-update` | "We *decided* something" | Templates on disk change for future scaffolds. | They form a funnel: **note → lesson once investigated → boilerplate convention at `seen: 3`.** --- ## Shared Step 0 — Apply this step's model tier (all three commands) ```bash source ~/.claude/skills/_lib/model-tier.sh apply_step_model lesson ``` Resolves this step's model from `.claude/model-mode.json` + `~/.claude/bank/model-tiers.md` (lesson never escalates — ceiling model even in god-mode), merges it into `.claude/settings.json`, and prints the restart notice. Informational only — continue regardless. Skip this step entirely when the cwd isn't a project (e.g. running from home). --- ## Command: `/lesson add <tag> <description>` 1. Normalize `<tag>` to lowercase, hyphen-separated. 2. Grep `~/.claude/bank/lessons/<tag>.md` (create it with the standard header if it doesn't