Change: Add proportionate security and evidence rule #14

Problem

Agent Handoff required risks and validation to be recorded but did not set a sufficiently strict threshold for security and evidence work to block acceptance or expand the current scope. A merely plausible or credible scenario could therefore create hardening, ADRs, gates, checkers, or separate stages before a useful vertical slice existed.

Scope

Publish Agent Handoff Standard 1.4 with a compact normative rule that:

• permits blocking or scope expansion only for a verified High or Critical current-scope risk;
• requires reproducible evidence from the current implementation or environment, or a directly applicable authoritative source;
• requires the affected asset or trust boundary, applicability, severity rationale, minimum sufficient control, and smallest verification;
• permits only a short, time-boxed investigation for a suspected High or Critical risk until it is confirmed or refuted;
• keeps Low, Medium, unrated, and unverified risks non-blocking;
• allows an exact acceptance criterion or verified legal or project requirement to block independently only when cited;
• preserves the existing security baseline and favors an early minimal end-to-end scenario;
• treats a 10–15% security-and-evidence share only as an optional planning heuristic that may be substantially exceeded only for a verified High or Critical risk or an exactly cited mandatory requirement.

Propagate the threshold through the standard, agent guide, handoff and reporting protocols, PR checklist, public adoption guidance, version metadata, release notes, project state, and the structural checker.

Non-goals

• No mandatory threat model for every Pull Request.
• No automated risk-severity scoring.
• No hard percentage limit.
• No new multi-stage approval gates.
• No Forge-specific policy.
• No change to SECURITY.md.
• No weakening of an established security baseline.

Acceptance criteria

• Blocking or scope-expanding security work requires a verified High or Critical current-scope risk.
• Verification requires reproducible evidence or a directly applicable authoritative source.
• A suspected High or Critical risk permits only a short, time-boxed investigation until confirmed.
• Low, Medium, unrated, and unverified risks cannot block an MVP or expand the current Issue.
• Exact acceptance criteria and verified mandatory requirements are cited when used as independent blockers.
• Reports distinguish verified blockers, time-boxed investigations, follow-up hardening, and owner-accepted risk.
• The 10–15% figure is explicitly non-binding.
• Existing security controls cannot be weakened without an explicit owner decision.
• The PR checklist contains one structural threshold item.
• The checker validates only the checklist structure, not evidence or severity.
• Standard version, citation metadata, active documentation, changelog, and release notes are synchronized to 1.4.
• GitHub Actions pass on the final HEAD.

Delivery

Use one short-lived branch and one Draft Pull Request. Do not merge without owner review.