iiRecord
Agentic AI Atlas · GAP-SEC-001: Governance Policy Layer
page:docs-harness-features-backlog-gaps-security-gap-sec-001a5c.ai
II.
Page reference

page:docs-harness-features-backlog-gaps-security-gap-sec-001

Reading · 1 min

GAP-SEC-001: Governance Policy Layer reference

Centralized policy engine for evaluating security rules at effect dispatch and task execution. Unified policy model replacing fragmented policy across breakpoint rules, env vars, and advisory hints.

Pagewiki/docs/harness-features-backlog/gaps/security/GAP-SEC-001.mdOutgoing · 0Incoming · 1

GAP-SEC-001: Governance Policy Layer

FieldValue
Categorysecurity
PriorityCritical
EffortL
StatusMissing

Description

Centralized policy engine for evaluating security rules at effect dispatch and task execution. Unified policy model replacing fragmented policy across breakpoint rules, env vars, and advisory hints.

Current State

Security policy fragmented: breakpoint rules in ~/.a5c/breakpoint-approvals/rules.json, advisory execution.permissions in task definitions, env var limits (BABYSITTER_MAX_ITERATIONS, BABYSITTER_HOOK_TIMEOUT). No unified evaluation point.

Target State

PolicyEngine evaluates declarative rules at effect dispatch and task execution. Policy types: rate-limit, permission, resource-limit, trust-level. Policy decisions persisted for audit trail. Existing breakpoint rules integrated as policy source.

Dependencies

  • None (foundation gap)

Key Files

ComponentPath
Breakpoint typespackages/sdk/src/breakpoints/types.ts
Breakpoint evaluatorpackages/sdk/src/breakpoints/evaluator.ts
Config modulepackages/sdk/src/config/
Process contextpackages/sdk/src/runtime/processContext.ts

Recommendation

Phase 1 implementation. Create packages/sdk/src/governance/ module. Define PolicyRule type with evaluation logic. Evaluate at effect dispatch in processContext.ts. Log decisions to structured JSONL.