Conversation
Document that workflow changes only take effect after merge due to GitHub's workflow validation security feature.
PR ReviewSummaryThis PR adds helpful CI documentation to ✅ Strengths
|
Greptile OverviewGreptile SummaryAdded documentation to
Confidence Score: 5/5
Important Files ChangedFile Analysis
Sequence DiagramsequenceDiagram
participant Dev as Developer
participant PR as Pull Request
participant GH as GitHub Actions
participant Main as Main Branch
Dev->>PR: Create PR with eval workflow
PR->>GH: Trigger workflow check
GH-->>PR: Security validation: Skip (new workflow)
Note over GH,PR: Workflow doesn't run on first PR
Dev->>Main: Merge PR to main
Main->>GH: Workflow now available
Dev->>PR: Create subsequent PR
PR->>GH: Trigger eval workflow
GH->>GH: Run 5 eval tests
GH-->>PR: Report results
|
- Use plugin_marketplaces and plugins inputs instead of claude_args - Add CRITICAL prompt section to prevent skill simulation - Install all 5 plugins from local marketplace
PR Review: Verify eval workflow runs on PROverviewThis PR updates the eval workflow configuration and documentation to clarify CI behavior and improve plugin installation approach. ✅ Strengths
🔍 Observations & Questions1. Workflow Trigger ScopeThe current
Recommendation: Add these paths to ensure the workflow validates its own changes: paths:
- 'hope/**'
- 'product/**'
- 'wordsmith/**'
- 'founder/**'
- 'career/**'
- 'eval/**'
- '.github/workflows/eval.yml' # Self-validation2. Missing CHANGELOG EntryPer CLAUDE.md:
This PR makes a breaking change to the workflow configuration (changing plugin installation method). Should add: ## [Unreleased]
### Changed
- **eval**: Workflow now uses `plugin_marketplaces` instead of `--plugin-dir` for more realistic testing
- **eval**: Added CRITICAL section to eval prompt requiring actual skill triggering
- **eval**: Documented GitHub workflow validation behavior in README3. Test CoverageWhat happens if:
Consider adding validation step before running tests. 🎯 Code Quality
🔒 SecurityNo security concerns. Changes are configuration-only and use existing GitHub Actions patterns. 📊 PerformanceNo performance impact. The plugin installation method change may be slightly slower (marketplace resolution) but more accurate for testing. ✏️ Suggested Changes
📝 Verdict🟡 SHIP + MONITOR Confidence: 85% Rationale:
Reversible: Yes - configuration change only Action Items:
|
Addresses the core bottleneck: AI output exceeds human verification capacity. presence skill: - Surfaces decisions made, assumptions, uncertainties, verification needed - Triggers on "lost track", "overwhelmed", "what just happened" - Frequency calibration based on context length - Integration with soul, gate, trace, shape verify skill: - Machine-verifiable acceptance criteria (moo principle #6) - Phase 1: Lock boolean criteria before building - Phase 2: Execute verification after building - agent-browser CLI integration for browser checks - Vague-to-boolean conversion table Both skills are [A]ctivation type - Claude knows these concepts but needs the structured framework to apply them consistently. https://claude.ai/code/session_01BE49Z9NJRgE43NSVPV4GYq
Purpose
Test that the eval workflow runs correctly now that it exists on main.
Changes
Expected