This release adds 5 notable features for engineering teams evaluating rollout.
Published 1mo
AI Agents & Assistants
✓ No known CVEs patched
✓ No known CVEs patched in this version
Summary
AI summaryArtifact validation now rejects markdown‑written .py files as goals, fixing silent success.
Full changelog
What changed
Suffering system hardening
- Stressor dedup now case-insensitive (fixed runaway accumulation bug from case variants)
MAX_STRESSORS = 5hard capforce_reset()breaks crisis loops after 3 consecutive cycles- Agents at cumulative load ≥ 0.90 no longer block goal execution — they work through it
Goal continuity (existence prompt)
- When a goal completes, outcome is written to
last_outcome.txtand injected into the next existence prompt asWHAT YOUR LAST GOAL PRODUCED - Recent
memory_setkeys surfaced in existence prompt so agents know what they've written - Recent goal history now shows
[DONE]/[FAILED]/[ABANDONED]tags
Artifact validation
.pyfiles now validated withcompile()— markdown written to a.pyfile no longer counts as a completed goalsynthesize_capabilityreturnsok: Falseon exec test failure instead of silently succeeding
Capability blacklisting
- Cross-cycle failure tracking: tools that fail once per goal run (never hit the within-run threshold) now accumulate in
broken_tools.jsonand get blacklisted at 5 total failures - Follow-on goal dedup is now fuzzy — >70% word overlap with a recent completed goal is rejected
Launcher
Double-click HollowAgentOS.bat. Requires Docker Desktop and Ollama with qwen3.5:9b.
Weekly OSS security release digest.
The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.
No spam, unsubscribe anytime.
Share this release
About Hollow
All releases →Related context
Related tools
Earlier breaking changes
- v5.7.32 Web dashboard removed; operator panel is canonical UI
Beta — feedback welcome: [email protected]