30 days ago
-
Sovereign-endpoint probe: boot wait + tok/s
-
v0.1.0
-
v0.1.0
-
v0.1.0
-
90s router window; None-guarded device check with audio-producing baseline
-
Posture evidence: one working entry suffices (continuous tool block logs once)
-
Fix posture-check window: epoch timestamps captured at speech-end/answer-end
-
Device-work check: luma OR posture evidence; epoch-time windows
-
Stream-resilient sampler + roleplay latency gates (20s/30s, LLM profile)
-
Fix IndentationError from the dedupe edit
-
video_state defined; dedupe; device-visual check repaired
-
Crash-proof sampler + audited-tool-call check
-
Fix video sampler (2-D luma, guards); export video_state for the device-work check
-
Grounded turn-2 keywords; shell tool env; workspace aligned
-
Device-work visibility check via laptop-region luminance
-
Tool-turn env fix: FREESTREAM_TOOLS=fs,shell + workspace path aligned with the bot sandbox
-
Tool turn: character reads seeded manifest via her terminal
-
Roleplay e2e: in-character conversation + policy gate + action checks
-
Fix NameError: restore chunk size in turn loop
-
Two-turn harness + transport-gap forensics (correct path)
-
Two-turn harness + transport-gap forensics; bot v12 render-pause knob
-
Win-condition ASR profile: tiny.en @1 thread
-
Whisper tiny @2 threads everywhere; bot forensics on failure
-
E2E win-condition harness: hf_hub_download fetch, speech-gated arrivals, join retry
-
Bounded answer wait + bot forensics dumps
-
MVP: end-to-end win-condition harness
-
Smoke test: verifies pushed repo artifacts (CLI exit codes, event chain, MCP handshake)
