The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
feat(commands): improve learn-eval with checklist-based holistic verdict
Tatsuya Shimomoto32e11b8701d4
diff --git a/commands/learn-eval.md b/commands/learn-eval.mdindex 18d8853cb..6f38894a9 100644--- a/commands/learn-eval.md+++ b/commands/learn-eval.md@@ -1,10 +1,10 @@----description: Extract reusable patterns from the session, self-evaluate quality before saving, and determine the right save location (Global vs Project).+description: "Extract reusable patterns from the session, self-evaluate quality before saving, and determine the right save location (Global vs Project)."---# /learn-eval - Extract, Evaluate, then Save-Extends `/learn` with a quality gate and save-location decision before writing any skill file.+Extends `/learn` with a quality gate, save-location decision, and knowledge-placement awareness before writing any skill file.## What to Extract@@ -51,36 +51,60 @@ origin: auto-extracted[Trigger conditions]```-5. **Self-evaluate before saving** using this rubric:+5. **Quality gate — Checklist + Holistic verdict**- | Dimension | 1 | 3 | 5 |- |-----------|---|---|---|- | Specificity | Abstract principles only, no code examples | Representative code example present | Rich examples covering all usage patterns |- | Actionability | Unclear what to do | Main steps are understandable | Immediately actionable, edge cases covered |- | Scope Fit | Too broad or too narrow | Mostly appropriate, some boundary ambiguity | Name, trigger, and content perfectly aligned |- | Non-redundancy | Nearly identical to another skill | Some overlap but unique perspective exists | Completely unique value |- | Coverage | Covers only a fraction of the target task | Main cases covered, common variants missing | Main cases, edge cases, and pitfalls covered |+ #### 5a. Required checklist (verify by actually reading files)- - Score each dimension 1–5- - If any dimension scores 1–2, improve the draft and re-score until all dimensions are ≥ 3- - Show the user the scores table and the final draft+ Execute **all** of the following before evaluating the draft:-6. Ask user to confirm:- - Show: proposed save path + scores table + final draft- - Wait for explicit confirmation before writing+ - [ ] Grep `~/.claude/skills/` by keyword to check for content overlap+ - [ ] Check MEMORY.md (both project and global) for overlap+ - [ ] Consider whether appending to an existing skill would suffice+ - [ ] Confirm this is a reusable pattern, not a one-off fix-7. Save to the determined location+ #### 5b. Holistic verdict-## Output Format for Step 5 (scores table)+ Synthesize the checklist results and draft quality, then choose **one** of the following:-| Dimension | Score | Rationale |-|-----------|-------|-----------|-| Specificity | N/5 | ... |-| Actionability | N/5 | ... |-| Scope Fit | N/5 | ... |-| Non-redundancy | N/5 | ... |-| Coverage | N/5 | ... |-| **Total** | **N/25** | |+ | Verdict | Meaning | Next Action |+ |---------|---------|-------------|+ | **Save** | Unique, specific, well-scoped | Proceed to Step 6 |+ | **Improve then Save** | Valuable but needs refinement | List improvements → revise → re-evaluate (once) |+ | **Absorb into [X]** | Should be appended to an existing skill | Show target skill and additions → Step 6 |+ | **Drop** | Trivial, redundant, or too abstract | Explain reasoning and stop |++ **Guideline dimensions** (informing the verdict, not scored):++ - **Specificity & Actionability**: Contains code examples or commands that are immediately usable+ - **Scope Fit**: Name, trigger conditions, and content are aligned and focused on a single pattern+ - **Uniqueness**: Provides value not covered by existing skills (informed by checklist results)+ - **Reusability**: Realistic trigger scenarios exist in future sessions++6. **Verdict-specific confirmation flow**++ - **Save**: Present save path + checklist results + 1-line verdict rationale + full draft → save after user confirmation+ - **Absorb into [X]**: Present target path + additions (diff format) + checklist results + verdict rationale → append after user confirmation+ - **Drop**: Show checklist results + reasoning only (no confirmation needed)++7. Save / Absorb to the determined location++## Output Format for Step 5++```+### Checklist+- [x] skills/ grep: no overlap (or: overlap found → details)+- [x] MEMORY.md: no overlap (or: overlap found → details)+- [x] Existing skill append: new file appropriate (or: should append to [X])+- [x] Reusability: confirmed (or: one-off → Drop)++### Verdict: Save / Improve then Save / Absorb into [X] / Drop++**Rationale:** (1-2 sentences explaining the verdict)+```++## Design Rationale++This version replaces the previous 5-dimension numeric scoring rubric (Specificity, Actionability, Scope Fit, Non-redundancy, Coverage scored 1-5) with a checklist-based holistic verdict system. Modern frontier models (Opus 4.6+) have strong contextual judgment — forcing rich qualitative signals into numeric scores loses nuance and can produce misleading totals. The holistic approach lets the model weigh all factors naturally, producing more accurate save/drop decisions while the explicit checklist ensures no critical check is skipped.## Notes@@ -88,4 +112,4 @@ origin: auto-extracted- Don't extract one-time issues (specific API outages, etc.)- Focus on patterns that will save time in future sessions- Keep skills focused — one pattern per skill-- If Coverage score is low, add related variants before saving+- When the verdict is Absorb, append to the existing skill rather than creating a new file