v0.5.0 — Multi-Run Review, OWASP Mapping, SARIF, GitHub Action

What's New in v0.5.0

The developer love release. Built to answer one question: how do you review AI agent work when you have multiple runs in flight?

Triage commands

• unworldly touched <session> — per-file risk summary ranked by danger. The fastest way to review what an agent changed.
• unworldly list --risk danger --agent "Claude Code" --since 2h — filter sessions instantly
• unworldly watch --tag pr-456 — label sessions for lookup later
• Sparkline ▁▂▃▄▅▆▇█ in unworldly list — see event density at a glance

Review workflow

• unworldly diff <a> <b> — compare two sessions: risk delta, file overlap, new commands
• unworldly review — interactive TUI like git add -p: approve/flag/skip/note each event, saves review.json
• unworldly snapshot before/after — git-aware diff of what the agent actually changed

Compliance & integrations

• unworldly report --owasp — first tool to map filesystem-level agent activity to OWASP Agentic AI Top 10 (2026)
• unworldly export --sarif — SARIF 2.1.0 export, appears in GitHub Security tab alongside CodeQL
• GitHub Action — posts session report as PR comment, uploads SARIF, optional fail-on-danger CI gate
• --on-danger — real-time webhook POST or desktop notification on any danger event

Install

pip install unworldly-recorder

Stats

• 6 new modules: owasp.py, diff.py, sarif.py, snapshot.py, alerts.py, review.py
• 210 tests passing
• Zero new dependencies