SPDD 2026-07-12: implement intent/authz enforcement, compliance fixtures, OTel Level 1 tests, and P2 spec hardening

[Copilot]

Addresses the SPDD daily plan covering 5 spec files (rotation indices 5–9). The critical P0 gap: AuthorizeTool policy enforcement was entirely unwired — all compiled policy constraints were advisory-only. P1 closes three missing compliance fixture gaps and converts 7 OTel Level 1 test stubs to passing tests. P2 consolidates scattered struct docs, adds structured safeguards, and normalizes concurrent-conflict behavior in the replace-label spec.

P0 — Intent authorization enforcement (pkg/intent/authz/)

New Authorizer type wired behind a FeatureEnabled feature flag (false = advisory for rollout safety):

a := authz.Authorizer{FeatureEnabled: true}

// DeniedTools checked first; AllowedTools nil=unrestricted, []string{}=deny-all
if err := a.AuthorizeTool(policy, toolName); err != nil { /* block */ }

// Scope ordering: none < feature_branch < any_branch
if err := a.AuthorizeWriteScope(policy, requestedScope); err != nil { /* block */ }

// Human approval gate and attempt cap
a.CheckHumanApproval(policy)
a.CheckAttemptLimit(policy, currentAttempts)

Also added PolicyCompiler.Compile() to pkg/intent/policy.go (was stub/advisory-only) with more-restrictive-wins merge semantics: DeniedTools union, AllowedTools intersection preserving nil vs empty distinction, HumanApproval OR, MaxAttempts min, WriteScope/Autonomy most restrictive. Unlinked and ambiguous PRs always produce safestDefaultPolicy() (propose-only, no writes, human approval required).

P1 — Compliance fixtures and OTel Level 1 tests

Three new intent-attribution compliance fixtures (specs/intent-attribution-compliance/), each with a dynamic Go test runner in pkg/intent/compliance_fixture_test.go:

• explicit-intent-wins.yaml — explicit ExplicitIntent metadata → source: explicit_metadata
• ambiguous-root-closing-issues.yaml — two distinct closing issues → status: ambiguous
• unlinked-pr-fail-closed.yaml — unlinked PR → propose_only, write_scope: none, human_approval_required: true

T-OT-001–007 converted from stubs to passing tests in pkg/workflow/otel_observability_formal_test.go covering: HTTPS endpoint emission, if-missing modes, multi-endpoint ordering, OTLP POST semantics, trace context injection (TRACEPARENT), /tmp/gh-aw/otel.jsonl mirror path, and header CSV format.

P2 — Spec hardening

specs/intent-attribution-agent-governance.md

• Added ## Entities section with normalized prose tables for all six Go structs (IntentRecord, ExecutionPolicy, PolicyRule, PolicyCondition, PolicyDecision, OutcomeRecord, EvidenceRecord) — previously only in scattered code blocks
• Sync Notes now records that .github/intent-policy.json does not exist in this repo (only objective-mapping.json does) with a concrete 5-item migration checklist
• AuthorizeTool implementation audit table updated from "Not wired" → "Wired (behind FeatureEnabled flag)" for AllowedTools, DeniedTools, WriteScope, HumanApprovalRequired, MaxAttempts; removed advisory-only risk note

specs/otel-observability-spec.md

• Added §15.5 Safeguards with three normative sub-sections each carrying testable acceptance criteria: OTLP export retry rate limits (max 3 retries, 30 s cap), PII scrubbing requirements (email, GitHub tokens, bearer headers), and cardinality budget enforcement (low-cardinality span names, 128-attribute cap, 2 KB value truncation)
• §17.1.1 heading updated from "Test ID Stubs" to "Level 1 Compliance Tests"; added Status: Implemented column

specs/replace-label-spec.md

• Added §5.9 Concurrent-Conflict Handling with four RFC 2119 requirements (RL-057–060): warning log on unexpected label set after successful PUT, no silent re-PUT after success, fail-closed on HTTP 422 with no auto-retry, SHOULD log before/after label state

specs/replace-label-compliance/

• New rl-004-concurrent-conflict.yaml with three scenarios: unexpected labels on success, no-retry enforcement, HTTP 422 fail-closed
• README.md fixture table updated with rl-004 row (RL-057–060)

[github-actions]

🤖 PR Triage

Field	Value
Category	feature
Risk	high
Priority Score	85/100
Score Breakdown	impact: 45, urgency: 25, quality: 15
Action	fast_track
OTel	@mnkeifer — contains OTel Level 1 tests

Large SPDD: intent/authz enforcement, compliance fixtures, OTel L1 tests, P2 spec hardening. 14 files, +2049/-35. Draft, CI unknown. High risk due to security/authz critical path. OTel content assigned to @mnkeifer.

Generated by 🔧 PR Triage Agent · 142.9 AIC · ⌖ 7.2 AIC · ⊞ 5.6K · ◷