Skip to content

Latest commit

 

History

History
31 lines (24 loc) · 2.94 KB

File metadata and controls

31 lines (24 loc) · 2.94 KB

Product Hardening Gap Closure

This page is the implementation map for the major gaps identified during the whole-project review. It is intentionally product-hardening oriented: every item below either adds a repo artifact that can be checked in CI, or records a non-goal boundary that prevents the verifier product from widening into a runtime system.

Closure Map

Gap Closure artifact Completion evidence
Root repo did not dogfood Agents Shipgate Root ../shipgate.yaml and advisory ../.github/workflows/agents-shipgate.yml agents-shipgate doctor --config shipgate.yaml --json; agents-shipgate scan -c shipgate.yaml --format json
Adoption benchmark was not a governance acceptance spec ../benchmark/agent-pr-governance/ defines the AgentPR Governance benchmark and case catalog; executable rows run real verify plus capability-lock diff assertions, while catalog-only rows remain backlog tests/test_governance_benchmark.py; tests/test_product_hardening_gap_closure.py validate the executable subset, catalog shape, category coverage, and decision coverage
Policy packs had matcher support but no test/explanation guidance policy-packs.md documents positive/negative fixture tests and release-decision contribution-rule explanations tests/test_policy_packs.py; tests/test_product_hardening_gap_closure.py
Trace/provenance did not have a unified local event and replay-bundle contract agent-workflow-evidence.md, agent-trace-event-schema.v0.1.json, and agent-workflow-evidence-bundle-schema.v0.1.json define no-chain-of-thought evidence shapes tests/test_product_hardening_gap_closure.py validates the schema contracts
Runtime inventory was only design-only and could blur the static boundary runtime-inventory.md records the opt-in-only boundary and prohibits use from default scan/verify tests/test_product_hardening_gap_closure.py checks the boundary wording

Product Invariants

  • report.json.release_decision.decision remains the only release gate.
  • merge_verdict and can_merge_without_human remain verifier/controller projections.
  • The default path remains local-first and static: no agent execution, no LLM calls, no MCP connections, no tool calls, no network, and no telemetry.
  • Governance benchmark, trace, and runtime-inventory artifacts do not introduce a report schema bump or a second verdict.

Ongoing Rule

When design-partner feedback exposes a false positive, missed capability, or unsafe-pass risk, add the smallest reproducible artifact to the governance benchmark catalog first. Add runtime execution, live inventory, or hosted workflow only behind an explicit opt-in command and after documenting its trust boundary.