Skip to content

v2.8.2 — Dual-Audit Consolidation: Handoff Graph Made Canonical + Bidirectionality Resolved

Choose a tag to compare

@Harery Harery released this 13 Jun 00:37
· 18 commits to main since this release

What is Praetor?

Praetor (Production Readiness, Audit, Evidence, Testing, Operations & Review) is a single-prompt, multi-agent orchestration framework that transforms any codebase into a comprehensive production readiness audit package. Paste one prompt into your LLM — 18 autonomous domain experts overseen by a 4-Judge Quality Council perform multi-disciplinary audits with 100% source-code traceability.


🚀 What is New in v2.8.2

Dual-Audit Consolidation

Two independent BALA7-30 audits of v2.8.1 — "opencode" (38 findings) and "Ro2a" (40 new findings). Every finding adjudicated against source before fixing: ~70 distinct valid findings fixed, 2 false positives confirmed. Both harnesses green.

Headline Changes

Theme What Changed
🔗 Canonical handoff registry HANDOFF_PROTOCOL now carries the complete directed-edge graph (~55 edges). Broadcast semantics, circular-pair initiation order, and sink-agent inbound pointers all explicit.
🔄 A00 ↔ A02 protocol coherence Orchestrator refusal condition exempts A02 uncorrectable distribution. Coverage Ledger write-misattribution fixed.
🎯 Precision & determinism Priority bands stated as independent windows. BLOCKED replaced with specific BLOCKED_BY_*. Snapshot gains AGENTS_NOT_ACTIVATED. QC failed reasons enumerated as closed set.
📊 Risk-register consolidation PHASE_6 §13: master view absorbs B.5 + E.4, dedups via RELATED_TO.
📖 Glossary + model tiers Added HANDOFF, MRR, Sev1/Sev2, SNAPSHOT_TOKEN, SNAPSHOT_DRIFT. Three-tier model-capacity guidance.
📝 Templates & mandates SLA bound to SLO placeholder. TC-ID label corrected. Burn-rate thresholds marked defaults. License "All rights reserved" removed.
🧪 Harness hardening Checker scans tests/ + tools/. New stage 11 for CI-pipeline copy parity. Fixture gains FLAW-11/12/13 (AWS key, postgres-creds URL, RSA key).

Also Includes (v2.8.0 + v2.8.1)

  • Tri-Audit Consolidation: ~75 findings from three independent audits
  • PRV↔E.3 column parity restored (10 columns, 1:1 mapping)
  • Self-consistency checker (11 stages)
  • Test fixtures with secret-scan harness
  • npm package: npx praetor-audit-kit
  • Claude Code skill: claude skill install Skill/praetor.skill

📦 Installation

# npm (recommended)
npx praetor-audit-kit

# Claude Code
claude skill install Skill/praetor.skill

# Manual — download the zip below, paste prompt into any long-context LLM

📊 By the Numbers

Count
Expert Agent Personas 18
Quality Council Judges 4
Execution Phases 7
Audience Categories 5
Register Types 12
Inter-Agent Protocols 13
Artifact Status Values 7 core + extended
ID Schemes 11

Full Changelog: CHANGELOG.md

One prompt. Eighteen experts. Six disciplines. Full traceability.

Full Changelog: v2.8.0...v2.8.2