The agent that produces work is never the agent that approves it. Generation and approval are separate passes so that output is checked against the original requirement — not against the producer's own claim that it is done.
Frameworks & Standards
Item
Value
Framework ID
independent-verification
Category
Operational
Version
1.0.0
Owner
Separation of Duties / Four-Eyes Principle (internal controls — COSO/ISACA)
Maturity
Established (governance) — applied to agent output in Maxim v1.3.3
You are applying the Independent Verification framework.
CONTEXT:
- Current task: [[task_description]]
- Domain: operational
- Stakeholders: [[stakeholder_roles]]
FRAMEWORK APPLICATION:
1. **Split generation from approval**: The producer and the verifier are distinct agents or passes.
2. **Verify against the requirement**: The independent reviewer re-checks against the requirement, not the producer's claim.
3. **Fail closed**: If no verifier is present and there is no explicit opt-in, refuse to mark the work done.
4. **Record the verdict**: Capture what was checked and the independent result.
OUTPUT STRUCTURE:
- Separation: Who produced vs. who verified
- Verification: What was checked against the requirement
- Gate Decision: Approved, rejected, or refused (fail-closed)
- Limitations: Any constraints or assumptions in the application
QUALITY CHECKS:
□ Producer and verifier are distinct
□ Verification was against the requirement, not the producer's claim
□ Absent a verifier and opt-in, the work was NOT marked done
□ Ethical considerations have been evaluated
Core Principles
Generation/approval split: Producer and verifier are distinct agents or passes.
Independent reviewer: Re-checks against the requirement, not the producer's claim.
Fail-closed gate: No verifier present and no opt-in → refuse to mark done.
Documentation: Record the independent verdict and what it was measured against.
orchestrator — enforces a separate verification gate in autonomous workflows
testing — proof-of-test re-validation as an independent pass
Testing Strategy
Validate that the verifier was a distinct agent/pass from the producer
Review one real example and one edge case (no verifier available → fail closed) before adopting the output
Confirm the verdict was measured against the requirement
Document adjustments made when the framework needed adaptation for context
Copyright (c) 2026 iSystematic Inc. Maxim is a product of iSystematic Inc. SPDX-License-Identifier: BSL-1.1 (Apache-2.0 after 4 years) See LICENSE at repo root. Framework definitions are reference material; value is delivered via Maxim's licensed runtime (pack-engine, MCP tools, dispatch, MemPalace).
1---2name: independent-verification3description: Independent Verification4---56# Independent Verification78## Purpose9The agent that produces work is never the agent that approves it. Generation and approval are separate passes so that output is checked against the original requirement — not against the producer's own claim that it is done.1011## Frameworks & Standards12| Item | Value |13|------|-------|14| Framework ID | `independent-verification` |15| Category | Operational |16| Version | 1.0.0 |17| Owner | Separation of Duties / Four-Eyes Principle (internal controls — COSO/ISACA) |18| Maturity | Established (governance) — applied to agent output in Maxim v1.3.3 |19| Primary References | Generation–approval separation; fail-closed verification gate |2021## Prompt Template22```23You are applying the Independent Verification framework.2425CONTEXT:26- Current task: [[task_description]]27- Domain: operational28- Stakeholders: [[stakeholder_roles]]2930FRAMEWORK APPLICATION:311. **Split generation from approval**: The producer and the verifier are distinct agents or passes.322. **Verify against the requirement**: The independent reviewer re-checks against the requirement, not the producer's claim.333. **Fail closed**: If no verifier is present and there is no explicit opt-in, refuse to mark the work done.344. **Record the verdict**: Capture what was checked and the independent result.3536OUTPUT STRUCTURE:37- Separation: Who produced vs. who verified38- Verification: What was checked against the requirement39- Gate Decision: Approved, rejected, or refused (fail-closed)40- Limitations: Any constraints or assumptions in the application4142QUALITY CHECKS:43□ Producer and verifier are distinct44□ Verification was against the requirement, not the producer's claim45□ Absent a verifier and opt-in, the work was NOT marked done46□ Ethical considerations have been evaluated47```4849## Core Principles50- **Generation/approval split**: Producer and verifier are distinct agents or passes.51- **Independent reviewer**: Re-checks against the requirement, not the producer's claim.52- **Fail-closed gate**: No verifier present and no opt-in → refuse to mark done.53- **Documentation**: Record the independent verdict and what it was measured against.5455## Applications & Use Cases56| Use Case | Application | Expected Outcome |57|----------|-------------|----------------|58| Independent review (Reviewer) | Independent verification pass | Output checked against the requirement |59| Proof-of-test (Tester) | Proof-of-test re-validation | Tests independently re-run, not trusted on claim |60| Regulated output (Security Analyst) | CSO independent check on regulated output | Compliance verified by a separate agent |61| Pre-tag audit (Pre-Release Audit) | Adversarial pre-tag verification | Release blocked until independently cleared |6263## Reference Materials64- [COSO Internal Control Framework](https://www.coso.org/) - Separation of Duties / Four-Eyes Principle65- [ISACA](https://www.isaca.org/) - internal controls guidance6667## Usage Guidelines68- **Start with context**: Identify the requirement the work will be verified against before generation.69- **Adapt, don't adopt**: Choose the verifier appropriate to the risk (reviewer, tester, CSO, pre-release audit).70- **Document decisions**: Record the independent verdict separately from the producer's output.71- **Review outcomes**: Confirm the verifier measured against the requirement, not the producer's claim.72- **Share learnings**: Feed recurring verification misses back into the review checklist.7374## Collaboration Protocol75- Apply independently unless a task explicitly requires another skill or framework76- Use structured handoff format: [Context] -> [Framework Applied] -> [Open Questions] -> [Next Action]7778## Ethical Guidelines79- ALWAYS keep the producer and the approver distinct80- NEVER let an agent self-certify regulated or high-risk output81- ALWAYS fail closed when no independent verifier is available8283## Success Metrics84- **Clarity**: The requirement being verified against is explicit85- **Consistency**: Similar output types get the same class of independent check86- **Stakeholder Alignment**: Four-eyes vocabulary improves cross-functional trust87- **Outcome Quality**: Fewer producer-self-certified defects reach release88- **Learning**: Verification misses generate stronger checklists8990## Related Skills91- `orchestrator` — enforces a separate verification gate in autonomous workflows92- `testing` — proof-of-test re-validation as an independent pass9394## Testing Strategy95- Validate that the verifier was a distinct agent/pass from the producer96- Review one real example and one edge case (no verifier available → fail closed) before adopting the output97- Confirm the verdict was measured against the requirement98- Document adjustments made when the framework needed adaptation for context99100---101<sub>Copyright (c) 2026 iSystematic Inc. Maxim is a product of iSystematic Inc. 102SPDX-License-Identifier: BSL-1.1 (Apache-2.0 after 4 years) 103See LICENSE at repo root. Framework definitions are reference material; value is delivered via Maxim's licensed runtime (pack-engine, MCP tools, dispatch, MemPalace).</sub>
Run npx skillmds@latest add drnabeelkhan/independent-verification in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.
Independent Verification It is listed under Coding & Dev Tools on SkillMD.
This skill has not completed SkillMD's automated safety review yet. Independent scanners report: SkillSpector: PASS, Skill Scanner: PASS. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.
This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.
Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.
DrNabeelKhan (@drnabeelkhan) published this skill. Their other Agent Skills are listed on their SkillMD profile.