← ClaudeAtlas

hv-learnlisted

Extract durable session learnings (gotchas, conventions, constraints) into .hv/KNOWLEDGE.md grouped by topic, and update the CLAUDE.md topic index. Use at end of a session that surfaced reusable knowledge, after a correction-rich debugging arc, or on "save what we learned", "capture this learning", "/hv-learn". Opus verification is on by default via learn.verify in config.json; set to false for fast/cheap mode.
l4ci/hv-skills · ★ 0 · AI & Automation · score 73
Install: claude install-skill l4ci/hv-skills
**Print the banner below verbatim before any other action — skip if dispatched as a subagent.** See `references/banner-preamble.md`. ``` ════════════════════════════════════════════════════════════════════════ 🧠 hv-learn · extract session learnings to KNOWLEDGE.md triggers: "learn this", "save gotcha" · pairs: hv-debug, hv-pause ════════════════════════════════════════════════════════════════════════ ``` # hv-learn — Capture Session Learnings ## Step 1 — Preflight ```bash .hv/bin/hv-preflight ``` See `docs/reference/preflight.md` for exit-code handling. **Initialize task list.** Follow the canonical pattern in `references/task-list-init.md` — load `TaskCreate(…)` via `ToolSearch select:TaskCreate,TaskUpdate` if needed, then create one task per phase below. Phases: 1. *Scan session* — transcript + recent commits sifted for durable gotchas (Step 2) 2. *Classify topic* — each candidate matched to a `KNOWLEDGE.md` topic (Steps 3–4) 3. *Merge into KNOWLEDGE.md* — entries appended under topic headings (Step 5) 4. *Update CLAUDE.md index* — `hv-managed-block knowledge` regenerates the managed block (Step 6) 5. *Verify (Opus)* — optional cold pass when `learn.verify: true` (Steps 7–8) 6. *Contradictions* — pending demotion candidates surfaced per-bullet at session end (Step 9) **Args parsing.** Before running the phases above, inspect the `args` value passed at invocation. If `args` contains any of the following flags, skip Steps 2–8 and jump directly to Step 1.5: