hipilot-skill-optimizer

name	hipilot-skill-optimizer
description	Optimize HiPilot skills based on HiTestBot results and failure patterns
triggers	["hipilot","skill","optimize","hitestbot","skill-creator"]
argument-hint	[skill-name]

HiPilot Skill Optimizer

Purpose

Analyze HiPilot skills against HiTestBot results to identify gaps, fix issues, and improve certification scores. This skill helps maintain high-quality skills that pass HiTestBot certification (6.0/6.0 target).

When to Activate

Use this skill when:

• HiTestBot reports failures (L1-L5)
• Skills need updating based on test results
• Creating new skills for HiPilot
• Refactoring existing skills for better clarity
• Fixing LEF order, QoR reporting, or tool switching issues

HiTestBot Scoring Reference

Layer	Score	What It Tests
L1	1.0	Claude responds to command
L2	1.0	Claude understands the task
L3	1.0	MCP tools are used correctly
L3b	1.0	Correct tool for each stage (dc_shell for synthesis, innovus for P&R)
L4	1.0	EDA tool executes without errors
L5	1.0	QoR metrics reported with explicit WNS/TNS numbers

Common Skill Issues & Fixes

1. L5 QoR Assessment Failure

Problem: No WNS/TNS numbers in Claude output Fix: Add explicit QoR reporting section:

### MANDATORY QoR REPORTING (L5 Requirement)
**CRITICAL: Report WNS/TNS in this exact format:**

WNS: 0.XXX ns TNS: 0.YYY ns

The test scorer looks for patterns:
- `WNS[:\s]*(-?[\d.]+)`
- `TNS[:\s]*(-?[\d.]+)`

2. L3b Process Validation Failure

Problem: Wrong tool for stage (e.g., innovus for synthesis) Fix: Explicitly state tool requirements:

**Tool:** dc_shell (REQUIRED - Innovus cannot synthesize RTL)

```javascript
// CORRECT - Stage 0 uses dc_shell
eda.start_tool({tool: "dc_shell", design_dir: process.env.HIPILOT_DESIGN_DIR})

// WRONG - Innovus cannot run synthesis
eda.start_tool({tool: "innovus"}) // ❌ Will fail

### 3. L4 EDA Execution Failure

**Problem**: IMPLF-53 LEF loading error
**Fix**: Document LEF order requirement:
```markdown
### LEF File Loading Order (CRITICAL)
Tech LEF (.tlef) MUST be loaded BEFORE cell LEFs (.lef):

```tcl
# CORRECT - Tech LEF first
set init_lef_file "sky130_fd_sc_hd.tlef sky130_fd_sc_hd_merged.lef"

# WRONG - Cell LEF before Tech LEF
set init_lef_file "sky130_fd_sc_hd_merged.lef sky130_fd_sc_hd.tlef" ;# ❌ IMPLF-53 error

## Skill Optimization Checklist

When reviewing a skill, verify:

- [ ] **L1/L2**: Clear purpose and intent recognition
- [ ] **L3**: Correct MCP tool usage patterns
- [ ] **L3b**: Correct tool specified for each stage
- [ ] **L4**: Error handling for common failures
- [ ] **L5**: Explicit WNS/TNS reporting requirements
- [ ] **Human-Like**: Incremental commands, not batch scripts
- [ ] **LittleBrain**: Uses knowledge tools for Tcl generation/QoR extraction

## Workflow

1. **Analyze HiTestBot Results**
   ```bash
   # Check latest test evidence
   ls test-evidence/latest/FLOW_REPORT.md

   # Identify failing layers
   grep -E "L[1-5].*:" test-evidence/latest/FLOW_REPORT.md

2. Find Affected Skills

   # Search skills for relevant content
   grep -r "wns\|tns\|QoR" skills/
   grep -r "dc_shell\|innovus" skills/
   

3. Update Skills

   • Add missing requirements
   • Strengthen error handling
   • Include exact patterns for test scoring

4. Re-run HiTestBot

   bin/hitestbot-eda /rtl2gds
   

5. Verify Fix

   • Check new FLOW_REPORT.md
   • Confirm score improved

Skill Template for HiTestBot Success

---
name: [workflow-name]
description: [Clear, specific description]
triggers: [relevant keywords]
---

# [Workflow Name]

## Purpose
[What this skill does and when to use it]

## Prerequisites
- [List required files/tools]
- [Any setup needed]

## Stage X: [Name] ([tool]) - ~[time] minutes
**Skip if:** `[checkpoint file exists]`

**Tool:** [dc_shell|innovus|pt_shell] (REQUIRED)

```javascript
// 1. Start correct tool
eda.start_tool({tool: "[tool]", design_dir: "$design_dir"})

// 2. Send Tcl
eda.send_tcl_nonblocking({tcl: stageX_tcl, description: "Stage X: [Name]"})

// 3. Wait for completion
eda.await_idle({timeout: [seconds], expected_tool: "[tool]"})

// 4. Check result
eda.get_last_result({lines: 50})

// 5. Run timing report for QoR
eda.send_tcl_nonblocking({tcl: "report_timing -max_paths 5", description: "Get timing"})
eda.await_idle({timeout: 30})
const output = await eda.get_last_result({lines: 50})

// 6. Report WNS/TNS (L5 REQUIREMENT)
console.log(`Stage X [Name]: WNS: [value] ns, TNS: [value] ns`)

Error Handling

Common Error: [error pattern] Fix: [specific solution]

Notes

• [Any additional context]

## References

- HiTestBot: `src/hitestbot/core/FlowCertifier.js`
- LittleBrain: `servers/knowledge/littlebrain/`
- Skills: `skills/`
- Test Evidence: `test-evidence/latest/`

## Examples

### Optimize ibex-rtl2gds-flow skill
```bash
# Check current score
cat test-evidence/latest/FLOW_REPORT.md | grep "L5"

# Result: L5 QoR Assessment: 0.0

# Fix: Add to skills/ibex-rtl2gds-flow.md
```markdown
### MANDATORY QoR REPORTING
After EVERY stage, report:

Stage X [Name]: WNS: 0.XXX ns, TNS: 0.YYY ns

Re-run test

bin/hitestbot-eda /rtl2gds

Verify: L5 QoR Assessment: 1.0

### Create new skill from HiTestBot failure
```bash
# Identify gap from test results
grep "blocking_stage" test-evidence/latest/flow_progress.json

# Create skill
/skill add fix-[specific-error]

# Populate with solution
# ...