rnd-learninglisted

# R&D Learning ## Overview RND agents discover non-obvious gotchas during iteration cycles. This skill defines how to capture them via the user's existing Learning Library and how to inject them into future builder prompts — so the same class of mistake is not repeated across tasks or sessions. ## When to Extract Only after a successful iteration cycle: ``` Build → Verify → NEEDS_ITERATION → Builder fixes → Re-verify → PASS ``` **NOT** on first-pass PASS (nothing was learned). **NOT** on FAIL without eventual resolution (nothing was fixed). ## What to Extract From the completed iteration cycle: - **Gotcha:** what failed — from the Verifier's feedback - **Fix:** what the Builder changed — from the iteration diff - **Combined learning:** what went wrong + how to avoid it next time Keep it terse. If you need a sentence to explain the fix, two bullets suffice. ## Format Match the existing Learning Library exactly: ```markdown ## Topic Name - Bullet describing the gotcha and its fix. Terse, factual, no prose. - Optional second bullet for context or edge cases. ``` ## Extension → Language File Mapping | Extensions | Learnings File | |---|---| | `.ts`, `.js`, `.jsx`, `.tsx`, `.mjs`, `.cjs` | `javascript.md` | | `.rs` | `rust.md` | | `.jl` | `julia.md` | | `.ex`, `.exs` | `elixir.md` | | `.gleam` | `gleam.md` | | `.sql` | `sql.md` | | `.css`, `.scss` | `css.md` | | `.sh`, `.bash` | `devops.md` | | `.md` (non-doc) | `devops.md` | | (fallback) | `devops.md` | ## Where to