Add code graph impact evidence integration

[safal207]

Goal

Add a ProofPath integration concept where code graph / blast-radius analysis becomes evidence for action-level security decisions.

Inspired by GitNexus-style code intelligence: before an AI agent performs a code, CI/CD, deploy, database, or infrastructure action, ProofPath should be able to consume impact evidence such as affected symbols, routes, modules, consumers, or risk level.

Core idea

AI agent proposes code or deploy action
  -> code graph / impact analyzer estimates blast radius
  -> ProofPath receives action context + impact evidence
  -> high-impact action without approval is held or blocked
  -> accepted/blocked decision is written to the audit log

Short line:

Code impact becomes action evidence.

Russian framing:

Влияние изменения в коде становится доказательством для решения о выполнении.

Why this matters

AI coding agents can make changes that look locally valid but have broad downstream impact:

• modify shared symbols;
• change public APIs;
• break route contracts;
• affect multiple consumers;
• change deployment or CI/CD behavior;
• alter database migrations;
• change security-sensitive modules;
• touch payment, auth, trading, or production code paths.

A valid GitHub token or CI credential does not mean the proposed code/deploy action is safe in context.

Proposed evidence shape

{
  "evidence_type": "code_impact",
  "risk_level": "HIGH",
  "affected_symbols": 12,
  "affected_routes": 4,
  "affected_consumers": 7,
  "touches_security_sensitive_code": true,
  "touches_production_deploy_path": true,
  "requires_maintainer_approval": true
}

ProofPath decision examples

LOW impact + reversible + scoped action -> ACCEPT
HIGH impact + no approval -> HOLD or BLOCK
CRITICAL impact + production path + no approval -> BLOCK
HIGH impact + explicit maintainer approval -> ACCEPT or AUDIT

Possible integration points

• GitNexus-style code graph output;
• static analysis tools;
• dependency graph analyzers;
• API route impact mapping;
• CI/CD diff classifiers;
• migration impact analyzers;
• security-sensitive path detectors.

Deliverables

• Add docs/code-impact-evidence.md.
• Define minimal code_impact evidence shape.
• Add examples for LOW/MEDIUM/HIGH/CRITICAL impact.
• Add a simulated demo under examples/code-impact-gate/.
• Add one safe low-impact change scenario returning ACCEPT.
• Add one high-impact shared-symbol change without approval returning BLOCK or HOLD.
• Add one approved high-impact change returning ACCEPT or AUDIT.
• Link the demo from the README execution-boundary demo matrix.
• Clarify that ProofPath consumes impact evidence but does not replace code review, tests, static analysis, CI, or deployment controls.

Acceptance criteria

• The demo is simulated and does not require a real GitNexus installation.
• The evidence format is implementation-neutral.
• The docs clearly separate impact analysis from enforcement.
• The project does not claim to solve code correctness by itself.
• The integration strengthens the full-lifecycle AI agent safety stack positioning.

Related

• Prepare ProofPath for external reviewers #32
• Add audit-log verification note and hash-chain checker #50
• Add policy configuration for scopes and approval rules #53
• Document full-lifecycle AI agent safety stack positioning #56