Integrate Product Compliance

Build compliance into the product loop without turning it into a late document gate. Agents may trace, propose, implement, test, and collect evidence; named humans retain interpretation, scope, risk acceptance, and assurance authority.

For ordinary security engineering without a confirmed control, contractual, or assessment scope, use the target's security practices and relevant engineering Skills directly.

1. Confirm Scope and Authority

Read the target's current ISMS or control system, policies, risk register, Statement of Applicability or equivalent, data and system inventories, contracts, assessment scope, prior findings, exceptions, and evidence store.

Confirm with named owners:

• applicable organization, products, locations, systems, data, suppliers, and contractual commitments
• exact standard, scheme, version, assessment objective, and assessment period
• control, risk, evidence, exception, release, and certification owners
• required assessor, auditor, QSA, legal, privacy, or security involvement

Do not infer applicability, certification status, TISAX objectives, PCI scope or SAQ eligibility, control interpretation, or acceptable risk. If these are unresolved, present options and stop the dependent branch.

Read references/frameworks.md when ISO/IEC 27001, TISAX, or PCI DSS applies. Re-check official sources because standards, assessment catalogs, FAQs, and effective dates change.

2. Classify the Product Change

Trace what changes across:

• data categories, flows, retention, residency, and deletion
• identities, roles, privileges, authentication, and administrative actions
• services, infrastructure, endpoints, payment pages, and trust boundaries
• suppliers, processors, libraries, agents, models, and external tools
• locations, teams, operating procedures, support, and incident response
• customer promises, assessment boundaries, and evidence obligations

Prefer scope reduction and simpler architecture when it preserves the intended outcome. A compliance label is not a reason to maximize controls or artifacts.

3. Map Risk, Controls, and Evidence

Use the target's existing identifiers and GRC system. For each affected risk or control record:

• requirement or obligation source and current version
• product or system behavior being protected
• implementation and control owner
• preventive, detective, and recovery mechanism
• deterministic test, policy gate, or manual review
• evidence source, provenance, retention, and access
• effectiveness signal and review trigger
• exception, expiry, compensating action, and risk owner when applicable

Do not copy copyrighted standards into the repository. Reference licensed sources and record only the target's interpretation, implementation, and evidence.

4. Encode Enforceable Controls

Turn a control into code only when its meaning, inputs, owner, and expected decision are stable enough to test. The named control owner retains the policy semantics. An agent may implement the approved rule; it must not translate an entire standard, contract, or policy into enforcement by inference.

For each executable control keep:

• the owned requirement, interpretation, control ID, version, and owner
• the structured input and the system or change boundary it covers
• the executable decision or invariant plus conforming and violating examples
• the enforcement point, failure behavior, exception path, and expiry
• the policy, data, engine, and test versions that produced the result
• the evidence destination and a check that the control runs where claimed

Use the target's existing schema validator, test framework, linter, CI gate, or platform control first. A small deterministic check is enough for a local invariant. Add a policy engine only when shared structured decisions, reuse, or runtime enforcement justify its operation and removal cost.

Machine-readable control and evidence formats may improve interchange and traceability when the target or assessor supports them. They do not prove applicability, effectiveness, or compliance by themselves.

5. Integrate the Product Loop

Use run-product-engineering and add compliance where it changes a decision:

Use threat modeling, abuse cases, data-flow review, architecture evidence, security testing, and assessor input only when the risk justifies them.

6. Generate Evidence from the System

Prefer evidence emitted by normal work:

• version control, reviews, signed or attributable changes
• CI/CD tests, policy results, provenance, and release records
• configuration, asset, access, vulnerability, and dependency state
• correlated audit events, traces, metrics, logs, and alerts
• incident, recovery, backup, restore, and continuity exercises
• supplier due diligence and contractual records
• training or manual evidence only where the control genuinely depends on it

Keep control mappings, implementation decisions, evidence references, and review history in version control when their sensitivity and owning system allow it. Generate audit views from those sources instead of maintaining a second Word, wiki, or ticket narrative by hand.

Git is an evidence source, not a substitute for a required assurance format. Use the current official TISAX/ISA, PCI ROC/AOC/SAQ, customer, assessor, or certification-body forms when the applicable route requires them, and link or generate them from the owned evidence where practical.

Evidence must identify scope, source, time, owner, version, and result. Protect it according to sensitivity and retention policy. An agent may assemble an evidence view but must not fabricate missing evidence or declare control effectiveness from file presence.

7. Apply Risk-Based Release and Exceptions

Use the target's existing risk classes. If none exist, propose a minimal model for human approval; do not silently impose a universal table.

Low-risk changes may use automated checks and ordinary review. Authentication, authorization, sensitive data, security logging, regulated payment flows, or assessment-scope changes usually require stronger targeted review, testing, evidence, rollback, and explicit release authority.

Emergency response may defer nonessential ceremony to restore safety or service. Preserve evidence, use the documented emergency authority, and complete post-change review and remediation.

Only the named risk owner may accept an exception. Record reason, scope, compensating controls, expiry, review trigger, and approval.

8. Review Effectiveness and Currentness

Feed control failures, incidents, false positives, audit findings, customer requirements, supplier changes, standard updates, and operational evidence back into risk, product, and harness decisions.

Agents may monitor evidence freshness, find gaps, propose remediation, prepare reviews, and execute approved bounded changes. Humans or qualified specialists retain:

• legal and contractual interpretation
• certification or assessment claims
• TISAX objectives and assessment scope
• PCI scoping, SAQ or ROC route, and QSA decisions
• risk acceptance and material release exceptions
• expansion from human-in-the-loop to human-on-the-loop

The integration is complete when scope and authority are explicit, affected risks and controls map to real product behavior, evidence arises from normal delivery, release follows the approved risk policy, effectiveness is observed, and no agent-generated claim exceeds verified assurance.

Related Skills

• run-product-engineering - operate the enclosing signal-to-outcome loop
• product-craft - shape trust, customer value, and viable bets
• scaffold-harness - assess governance and the operating level
• coding-discipline and completion-gate - implement and verify controls
• build-autonomous-agents - constrain agent runtime data, tools, and effects
• manage-infrastructure-as-code - plan, policy-check, apply, and reconcile infrastructure desired state
• documentation-and-adrs - record accepted consequential trade-offs

Sources

• NIST OSCAL - machine-readable control, implementation, assessment, and result models
• Open Policy Agent in CI/CD - executable policy checks for configuration and infrastructure data