Implement ResolveRisk and Authorizer.AuthorizeTool in pkg/intent

[Copilot]

specs/intent-attribution-agent-governance.md specifies a ResolveRisk risk-classification function and an Authorizer.AuthorizeTool enforcement point, but neither existed in pkg/intent — PolicyCompiler.Compile remained purely advisory with no tool-level gate.

Risk classification

• Added Domains []string, Priority string, Risk string fields to IntentRecord (kept separate from Labels, which PolicyCondition still matches against).
• Added ResolveRisk(rec IntentRecord) string in new pkg/intent/governance.go: explicit Risk always wins; otherwise derives from domains/priority (security+critical or production → high, infrastructure → medium, documentation → low, else unknown).

Tool authorization

• Added Authorizer struct with AuthorizeTool(policy ExecutionPolicy, tool string) error:
  • DeniedTools always wins, even if the tool also appears in AllowedTools (ErrToolDenied).
  • nil AllowedTools means unrestricted; a non-nil, empty AllowedTools denies every tool — this nil-vs-empty distinction had to be preserved rather than a naive slices.Contains check (ErrToolNotAllowed).

policy := intent.ExecutionPolicy{AllowedTools: []string{"read"}, DeniedTools: []string{"write"}}
intent.Authorizer{}.AuthorizeTool(policy, "write") // ErrToolDenied
intent.Authorizer{}.AuthorizeTool(policy, "exec")  // ErrToolNotAllowed
intent.Authorizer{}.AuthorizeTool(policy, "read")  // nil

Tests & docs

• Added pkg/intent/governance_formal_test.go covering the full behavioral coverage map from the issue (explicit-override precedence, each risk tier, deny/allow-list semantics, fail-closed compilation for unlinked/ambiguous, and edge cases for empty/nil inputs).
• Updated pkg/intent/README.md with the new public API and clarified PolicyCondition matches IntentRecord.Labels, not the new dedicated fields.
• Updated the spec's Authorizer.AuthorizeTool Implementation Audit to reflect the implementation status; wiring it into the orchestrator's execution path remains a follow-up, as noted in the spec itself.

• Fixes [formal-spec] intent-attribution-agent-governance.md — Formal model & test suite — 2026-08-18 #53741

[Copilot]

Pull request overview

Implements intent risk classification and tool authorization APIs, with formal tests and documentation updates.

Changes:

• Adds risk metadata and deterministic ResolveRisk.
• Adds deny/allow-list enforcement via Authorizer.
• Adds formal coverage and API documentation.

Show a summary per file

File	Description
pkg/intent/governance.go	Implements risk resolution and authorization.
pkg/intent/governance_formal_test.go	Tests governance invariants and edge cases.
pkg/intent/resolver.go	Adds classification fields to intent records.
pkg/intent/README.md	Documents the new APIs.
specs/intent-attribution-agent-governance.md	Updates implementation status.

Review details

💡 Add a code-review agent skill for context-aware, tailored reviews. Learn more in the docs.

• Files reviewed: 5/5 changed files
• Comments generated: 2
• Review effort level: Balanced

[github-actions]

🎉 This pull request is included in a new release.

Release: v0.87.1