Skip to content
Back to skills

Verify

ASecurity

Run deterministic static verifier scripts and blinded multi-persona rubrics to prove requirements. Trigger via /verify.

  • 2 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added September 3, 2026
ai-agentspythongonodedatabase

Security analysis

A100/100

Pro scans all 16 files and shows the line behind each finding

Scanned September 20, 2026

npx -y skills add ksprashu/agent-skill-forge --skill verify --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Verify?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Verify
[![Security: A β€” Skills Directory](https://www.skillsdirectory.com/api/skills/ksprashu-verify/badge)](https://www.skillsdirectory.com/skills/ksprashu-verify)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: verify
description: Run deterministic static verifier scripts and blinded multi-persona rubrics to prove requirements. Trigger via /verify.
---

# Verify: Dual-Layer Expectation Verification

Enforce Expectation-Grounded Alignment (EGA) through deterministic static check scripts and blinded dynamic judge rubrics.

---

## 🎯 Goal
Prove that deliverables satisfy architectural contracts, word bounds, schemas, and functional behavior with zero author bias.

---

## πŸ“‹ Step-by-Step Workflow

1. **Synthesize Expectations First**: Before writing deliverables, define deterministic static check scripts (`verify_static.py`) and blinded rubrics (`rubric.json`).
2. **Execute Static Layer**: Run the Python/Node static verifier to check AST, schema validity, link integrity, and formatting bounds.
3. **Run Blinded Dynamic Judge**: Evaluate the raw deliverable without exposing the author's internal thinking or conversation history.
4. **Conduct Forensic Integrity Audit (Anti-Mock Gate)**:
   - Inspect test suites, fixtures, and source files to verify real implementation.
   - Assert zero dummy mock facades replacing core business logic.
   - Assert zero hardcoded outputs designed specifically to satisfy unit tests.
   - Assert zero deleted, weakened, or commented-out test assertions.
   - If circumvention is detected, issue an immediate **Binary Veto** halting progression.
5. **Enforce Doubt-Driven Disproof**: The judge actively searches for edge-case failures, broken assumptions, and unhandled errors.
6. **Execute Independent Victory Audit**:
   - Before final sign-off, run a clean-slate empirical test and build pass with zero shared context from implementing agents.
   - Emit formal verdict: `VICTORY CONFIRMED` or `VICTORY REJECTED`.
7. **Remediate on Failure**: If any check fails, generate a concrete delta report and retry (max 3 retries).

---

## πŸ’‘ Concrete Examples

### 1. Fixture: Static Verifier Script (`verify_static.py`)
```python
import os, sys, json

def verify():
    # 1. File exists
    assert os.path.exists("dist/bundle.json"), "Missing dist/bundle.json"
    
    # 2. Schema validity
    with open("dist/bundle.json") as f:
        data = json.load(f)
    assert "version" in data, "Missing version field in bundle.json"
    assert len(data.get("items", [])) > 0, "Items list cannot be empty"
    
    print("βœ… Static verification PASSED")

if __name__ == "__main__":
    verify()
```

### 2. Fixture: Forensic Integrity Audit Checklist
```markdown
# Forensic Integrity Audit Check
- [ ] No hardcoded mock returns replacing business logic
- [ ] Real database / network integration verified (no fake in-memory facades)
- [ ] No weakened or deleted test assertions from baseline
- [ ] No `@ts-ignore` or file-wide linter disable comments hiding defects
- [ ] Verdict: PASS or BINARY VETO
```

---

## 🚫 Hard Constraints

*   **NEVER** sign off on deliverables without running the deterministic static verification script.
*   **NEVER** pass author rationale or conversational history to the blinded dynamic judge.
*   **NEVER** tolerate mock facades, cheated assertions, or hardcoded return values (Binary Veto).
*   **NEVER** report task completion without clean-slate Victory Audit confirmation.
*   **NEVER** allow more than 3 remediation retry cycles without halting for human inspection.

Files in this skill

  • README.md466 B
  • SKILL.md2.4 KB
  • references/dynamic_judges.md6.7 KB
  • references/ega_paradigm.md5.9 KB
  • references/knowledge_grounding.md3.6 KB
  • references/model_calibration.md5.1 KB
  • references/persona_roles.md4.3 KB
  • references/rsi_and_closure_protocol.md3.8 KB
  • references/static_verifiers.md4.8 KB
  • scripts/compile_ega_harness.py19.9 KB
  • scripts/ega_closure_engine.py5.8 KB
  • scripts/ega_controller.py5.2 KB
  • scripts/ega_loop_runner.py12.5 KB
  • scripts/mine_brain_transcripts.py3.9 KB
  • scripts/rsi_dream_engine.py6.5 KB
  • scripts/verify_okf.py3.2 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.