-
Fix model resolution bug for non-Ollama models in easy API (v3.0.1)
-
clean: remove stale reports and results files"
-
README: complete rewrite with visual architecture, animations, feature map, usage guide
-
clean: remove temp scripts, patches, and dev artifacts before public release
-
Update README with interactive hero, mermaid tracks, and v3 features
-
v3.0.0 Production Release: Hardened framework, strict tool validation, test suite robustification
-
Add fix: apply_validate_fix.py patches mock heuristic detection
-
Add critical fix: apply_llm_timeout.py patches OpenAI backend with 60s timeout
-
Add local_test_with_api.sh — full test suite including cloud model tests
-
Add apply_fixes.py — applies all 5 test fixes locally before testing
-
Fix Issue 3: Add 60s default timeout to OpenAI client
-
Fix Issue 3: Add 60s default timeout to OpenAI-compatible backend
-
Fix Issue 4: Make mock heuristic detection more resilient
-
Fix Issue 3: Add 60s timeout to API calls in prod_test.py
-
Fix Issue 2: Make backpressure test T1.6 more robust with consumer_started sync
-
Document all 5 test fixes for v3.0.0
-
Fix Issue 1: Make Trajectory None guards thorough (matching SRE patch quality)
-
Upload build_and_publish.py
-
Upload run_all_tests.sh
-
v3.0.0 — update version from 3.0.0a1 to 3.0.0
-
v3.0.0 — update version from 3.0.0a1 to 3.0.0
-
SRE: regression test for all 5 vulnerability patches
-
SRE: auto-apply patches on import (add sre_patches import to __init__)
-
SRE fixes: 5 critical vulnerability patches (dict snapshot, UNKNOWN reject, token cap, fine-grained lock, None guard)
-
harden: tests for null safety, timeouts, validation, graceful degradation
-
harden: defensive core — null safety, timeouts, graceful degradation, domain-agnostic
-
gradual-adoption: export state_delta, falsification_critic, sandbox_hooks from __init__
-
gradual-adoption: wire state_delta + falsification_critic into Orchestrator (backward-compat)
-
first-principles: tests for state-delta, falsification critic, sandbox hooks
-
first-principles: sandbox_hooks.py — PEP 578 kernel-level audit hooks (unbypassable)
-
first-principles: falsification_critic.py — Popperian scoring by adversarial assertions
-
first-principles: state_delta.py — O(1) Markovian critic (no history needed)
-
Sprint 12: CHANGELOG.md + v3 migration guide
-
Sprint 11: benchmark_v3.py — robustness + competitor-style benchmark suite
-
Track D tests: fingerprint, dataset, prompt_pack, shadow_eval, optimizer, distillation
-
Sprint 10C: lora_plan.py + distill.py — adapter/distillation planning (dry-run, no GPU required)
-
Sprint 10B: optimizer.py — agentic decision policy (improving/plateau/degrading/rollback)
-
Sprint 10A: shadow_eval.py — compare candidate vs baseline safely
-
Sprint 9C: prompt_pack.py — epigenetic optimization (prompts before weights)
-
Sprint 9B: dataset.py — trace → filtered training dataset builder
-
Sprint 9A: fingerprint.py — capability fingerprint from execution traces
-
Sprint 9-10: optimization package init
-
Sprint 5-7 tests: routing, MAS generator, skills
-
Sprint 7: skills/ci.py — skill testing, promotion, rollback, immune scan
-
Sprint 7: skills/schema.py — SkillCard + SkillGenome with versioning
-
Sprint 7: skills package — SkillCard schema, runtime, CI, export
-
Sprint 6: mas_generator.py — use-case → generated multi-agent system
-
Sprint 5: routing.py — LLMCallRouter, ModelSelector, cost homeostasis
-
Sprint 4+8: protocol and quorum tests
-
Sprint 8: quorum.py — consensus/disagreement topology switches + critic ensemble
