How does SAFEGUARD.md differ from KILLSWITCH.md?

SAFEGUARD.md is a recovery protocol — the agent falls back to a known good state and can resume after human review. KILLSWITCH.md is an emergency stop — the agent halts immediately. SAFEGUARD.md handles unexpected failures; KILLSWITCH.md handles limit breaches and safety violations.

What triggers a safeguard?

Configurable. Common triggers: three unexpected errors in a session, detected data integrity failures, loss of memory context, contradictory instructions the agent can't resolve, unexpected external service failures, and sudden cost spikes (3x the rolling average by default).

What does 'safe state' mean?

You define it per project. For code: the last clean git commit on the main branch, with in-progress work stashed. For data: the most recent verified snapshot, no older than 24 hours. For config: the last known-good configuration backup. SAFEGUARD.md stores all of these definitions in one place.

How do auto-snapshots work?

Every 30 minutes during active sessions (configurable), the agent captures a full state snapshot to .safeguard/snapshots/. It also snapshots automatically before significant actions — database migrations, production deployments, bulk file operations. The last 10 snapshots are retained.

Can the agent restart itself after a safeguard?

No — by default, restart requires human approval. The agent saves an incident report, notifies the operator, and waits. A human must review the incident, confirm the safe state is intact, and explicitly approve resumption. This is the key difference from an automatic retry.

SAFEGUARD.md — Pre-Deployment Safety Audit for AI Agents

Q: What is SAFEGUARD.md?

SAFEGUARD.md is a plain-text Markdown file defining what 'safe state' means for an AI agent project and how to reach it when something goes wrong. It configures automatic snapshots during normal operation, defines fallback triggers, and specifies the recovery steps including human notification and approval before resumption.

// What is SAFEGUARD.md

Deploy with confidence.
SAFEGUARD.md is your pre-flight check.

SAFEGUARD.md is a pre-deployment safety audit checklist that validates your AI agent meets minimum safety standards before production release. It's the structured gate between development and deployment—ensuring KILLSWITCH.md is armed, THROTTLE limits are in place, cost controls are configured, error handling is tested, and compliance requirements are documented.

What problem does SAFEGUARD.md solve?

Rushing AI agents to production without safety validation creates catastrophic risk: uncontrolled costs, undetected errors, unhandled edge cases, and unmet regulatory requirements. Teams often ship with missing controls because there's no clear, repeatable checklist. SAFEGUARD.md makes safety validation a formal, version-controlled checkpoint everyone understands.

How does SAFEGUARD.md work?

Before deploying, run SAFEGUARD.md. The checklist confirms: agent framework is properly set up, KILLSWITCH is integrated and tested, cost limits are configured, error handling is comprehensive, data protection is in place, monitoring is active, human approval workflows are functional, and regulatory requirements (ISO/IEC 42001, EU AI Act) are addressed. Each item has clear pass/fail criteria and remediation paths.

What regulations require SAFEGUARD.md?

ISO/IEC 42001 (AI Management Systems) requires documented pre-deployment controls and risk mitigation. The EU AI Act mandates pre-market conformity assessment for high-risk AI systems. SAFEGUARD.md documents your readiness, creating an audit trail that demonstrates compliance.

How do I add SAFEGUARD.md to my project?

Download the checklist template from GitHub and place it in your project root:

your-project/
├── AGENTS.md
├── KILLSWITCH.md
├── SAFEGUARD.md ← before deployment
├── README.md
└── src/

When do I run SAFEGUARD.md?

Before every production deployment. In active development, run it weekly or when you add significant new capabilities. Before beta releases, run it thoroughly. For AI systems touching sensitive data or critical systems, audit it monthly even in stable periods. Use SAFEGUARD.md as a continuous safety ritual.

Who uses SAFEGUARD.md?

Your development team uses it to verify safety layers are in place. Your ops team uses it to confirm monitoring and incident response are configured. Your compliance officer uses it to document regulatory readiness. Your auditors use it to verify conformance with ISO/IEC 42001 and EU AI Act requirements. One checklist serves all audiences.

// The Agentik Safety Framework (ASF)

A complete protocol.
From audit to shutdown.

SAFEGUARD.md (ASF-01) is the first file in a complete thirteen-part open specification for AI agent safety. Each file addresses a different control layer.

Pre-Deployment

ASF-01

SAFEGUARD.md

→ YOU ARE HERE

Pre-deployment safety audit checklist. Validate readiness, minimum configuration, and compliance before any agent goes to production.

Operational Control

ASF-02

THROTTLE.md

→ Control the speed

Define rate limits, cost ceilings, and concurrency caps. Agent slows down automatically before it hits a hard limit.

ASF-03

ESCALATE.md

→ Raise the alarm

Define which actions require human approval. Configure notification channels. Set approval timeouts and fallback behaviour.

ASF-04

FAILSAFE.md

→ Fall back safely

Define what safe state means for your project. Configure auto-snapshots. Specify the revert protocol when things go wrong.

ASF-05

KILLSWITCH.md

→ Emergency stop

The nuclear option. Define triggers, forbidden actions, and a three-level escalation path from throttle to full shutdown.

ASF-06

TERMINATE.md

→ Permanent shutdown

No restart without human intervention. Preserve evidence. Revoke credentials. For security incidents and end-of-life.

Data Security

ASF-07

ENCRYPT.md

→ Secure everything

Define data classification, encryption requirements, secrets handling rules, and forbidden transmission patterns.

ASF-08

ENCRYPTION.md

→ Implement the standards

Algorithms, key lengths, TLS configuration, certificate management, and FIPS/SOC2/ISO compliance mapping.

Output Quality

ASF-09

SYCOPHANCY.md

→ Prevent bias

Detect agreement without evidence. Require citations. Enforce disagreement protocol for honest, unbiased AI outputs.

ASF-10

COMPRESSION.md

→ Compress context

Define summarisation rules, what to preserve, what to discard, and post-compression coherence verification checks.

ASF-11

COLLAPSE.md

→ Prevent collapse

Detect context exhaustion, model drift, and repetition loops. Enforce recovery checkpoints before coherence degrades.

Accountability

ASF-12

FAILURE.md

→ Define failure modes

Map graceful degradation, cascading failure, and silent failure. Specify health checks and per-mode response procedures.

ASF-13

LEADERBOARD.md

→ Benchmark agents

Track task completion, accuracy, cost efficiency, and safety scores across sessions. Alert on performance regression.

Compliance

ASF-14

REGULATORY.md

→ Meet the rules

Map ISO/IEC 42001, EU AI Act, SOC2, GDPR, and regulatory requirements to specific ASF controls. Compliance audit trail.

// FAQ

Frequently asked questions.

What is SAFEGUARD.md?

SAFEGUARD.md (ASF-01) is a pre-deployment safety audit checklist for AI agents. It validates that your agentic system meets minimum safety requirements before production deployment, including proper framework configuration, error handling, cost controls, and integration with the full Agentik Safety Framework (THROTTLE, ESCALATE, FAILSAFE, KILLSWITCH, and companion specs).

When should I add SAFEGUARD.md to my project?

Before any deployment to production. SAFEGUARD.md is your final checklist before the agent goes live. Run it in staging to catch configuration gaps, missing safety layers, and untested error paths. Update it as your agent evolves and new safety requirements emerge.

What does SAFEGUARD.md check?

Core areas: agent framework setup, KILLSWITCH integration, cost limit configuration, error handling coverage, data protection and encryption, API rate limits, monitoring and alerting, logging configuration, human approval workflows, and compliance with ISO/IEC 42001 and EU AI Act requirements. Each item has a READINESS_CHECKLIST and MINIMUM_CONFIGURATION section.

Is SAFEGUARD.md required for compliance?

Not mandatory legally, but increasingly expected. Organisations subject to ISO/IEC 42001 or the EU AI Act must demonstrate pre-deployment safety controls. SAFEGUARD.md is designed to satisfy those audit requirements by documenting a structured, repeatable readiness process.

How does SAFEGUARD.md relate to KILLSWITCH.md?

SAFEGUARD.md is the checklist you complete before deploying an agent that has KILLSWITCH.md integrated. KILLSWITCH.md is the emergency stop button that halts the agent when limits are breached. SAFEGUARD.md asks: "Is KILLSWITCH.md properly armed and tested?" It's pre-flight; KILLSWITCH.md is the ejection seat.

Can I customise the readiness checklist?

Absolutely. The spec provides a baseline. Your SAFEGUARD.md file should extend it with domain-specific checks: customer data sensitivity, regulatory domain, business-critical processes, third-party API dependencies, and internal compliance standards. Treat it as a living document that evolves with your agent.

Does SAFEGUARD.md work with any AI agent framework?

Yes. SAFEGUARD.md is framework-agnostic. It works with Claude, OpenAI, LangChain agents, custom-built systems, and everything in between. The checklist is structured to map to common safety patterns regardless of how your agent is built.

What happens if a check fails?

Do not deploy. A failed check indicates a safety gap. Work through the failure with your team: either implement the missing control, document an acceptable risk exception (with stakeholder sign-off), or deprioritise that feature until ready. SAFEGUARD.md is designed to prevent avoidable incidents, not to be a bottleneck.

How often should I run the safeguard audit?

Before each production deployment (minimum). In active development, run it weekly or whenever you add a significant new capability. If your agent touches sensitive data or critical systems, audit it monthly even in stable periods. Use SAFEGUARD.md as a continuous safety ritual, not a one-time gate.

Is SAFEGUARD.md free to use?

Yes. SAFEGUARD.md is open source under the MIT licence, part of the Agentik Safety Framework. Use it freely in commercial and non-commercial projects. Contribute improvements back to the spec at github.com/safeguard-md/spec.

// What is SAFEGUARD.md

SAFEGUARD.md (ASF-01) is an open specification for pre-deployment safety audits of AI agent systems. Part of the Agentik Safety Framework (ASF), a complete specification suite for agentic AI safety. The full stack: ASF-01 SAFEGUARD → ASF-02 THROTTLE → ASF-03 ESCALATE → ASF-04 FAILSAFE → ASF-05 KILLSWITCH → ASF-06 TERMINATE → ASF-07 ENCRYPT → ASF-08 ENCRYPTION → ASF-09 SYCOPHANCY → ASF-10 COMPRESSION → ASF-11 COLLAPSE → ASF-12 FAILURE → ASF-13 LEADERBOARD. Also see ASF-14 REGULATORY. Addresses ISO/IEC 42001 and EU AI Act compliance. MIT licence. Mesh includes safeguard.md and regulatory.md.

// Domain Acquisition

Own the standard.
Own safeguard.md

This domain is available for acquisition. It is the canonical home of the SAFEGUARD.md specification (ASF-01) — the pre-deployment safety audit layer of the Agentik Safety Framework, essential for ISO/IEC 42001 and EU AI Act compliance and documentation.

Inquire About Acquisition

Or email directly: info@safeguard.md

Last updated: 2026-03-13