v3.8.0 — Skills Expansion + Validation Theater Guard
What's New
7 New Skills
- /snr — Signal/Noise Ratio: 6 modes (daily triage, meeting gate, decision memo, anti-distortion, agent gate, weekly audit)
- /sci-hypothesis — Computational Polymath: Banality Filter, Novelty Pressure ≥8, Discovery Score formula
- /plan-review — Human-in-the-loop plan annotation (adapted from Plannotator ⭐4776). New: Kill Condition field
- /changelog-gen — git commits → user-facing release notes
- /lead-research — OSINT due diligence: theHarvester + Subfinder + Nuclei
- /meeting-insights — transcript → structured wiki entry
- /harvest — byproduct asset scouting from failed/unfinished projects
New Hook: Validation Theater Guard
Detects when agents create validators and run them on synthetic data in the same session:
- Flags
create_synthetic_dataset/mock_datain validator files - Alerts on F1=1.000 / perfect scores in Bash output
- Prompts Zero-Based questions before SUCCESS declaration
Evidence Marker Split (integrity.md)
[VERIFIED-REAL]— real-world data with sources (required for hypothesis validation)[VERIFIED-SYNTHETIC]— synthetic/mock (unit tests only)[VERIFIED-INLINE]— quick sanity check
CI & Infrastructure
- plugin.json added to all 39 extensions (fixes smoke test failures)
FORCE_JAVASCRIPT_ACTIONS_TO_NODE24=true— ready for June 2 GitHub Actions migration- install.sh:
sync_global_skills()— all skills available globally after install - AGENTS.md: project agent rules (Read First protocol, skill routing, cannot-do list)
Stats
- Tests: 1093 passing
- Coverage: 86% (local/Windows) · 65% (CI/Linux)
- Skills: 40 total (10 core + 30 extensions)
- Hooks: 50 deterministic guards