Skip to content

Shipwright v2.0.0

Latest

Choose a tag to compare

@EdgeCaser EdgeCaser released this 02 Apr 22:08
· 103 commits to main since this release

What's new in v2.0.0

Benchmark harness + proof framework

Six scored scenarios with a fixture corpus and blind-review pipeline. Full v2 proof methodology documented in docs/shipwright-v2-proof-method.md, shipwright-v2-proof-runbook.md, shipwright-v2-technical-spec.md, and shipwright-v2-strategy-brief.md.

Deterministic integration layer

classify-request, format-facts, and pricing-diff scripts decouple structured fact extraction from LLM judgment. Source adapters for PyPI and crates.io. Collector-first research with bounded escalation and cache.

Red-team adversarial review

New /challenge skill with rubric, golden output, and recovery playbook for DEFEND/ESCALATE verdicts.

Slack mention agent

Full TypeScript Socket Mode agent (slack-agent/) — command allowlists, thread listening with TTL, per-thread session continuity. No public URL required.

Full test suite

13 test files covering all scripts end-to-end. CI/CD via GitHub Actions (validate + Claude code review workflows).

Artifact schemas + validation

JSON schemas for PRD, Strategy, and Challenge Report. validate-artifact.mjs enforces structure at output time.

Hardened skills + agents

All 42 skills updated with depth enforcement, evidence bar, and Shipwright signature. Agent contracts extended with explicit handoffs and failure remediation (AGENTS.md).


Upgrade from v1.x: No breaking changes to skill invocation. New scripts require Node 18+ (npm install in root). Slack agent requires separate setup — see slack-agent/README.md.