Skip to content

TaskChad OS v1.2.0 — Talk Mode Voice Co-Founder

Choose a tag to compare

@TheSmokeDev TheSmokeDev released this 02 Aug 20:54
· 10 commits to master since this release

The complete voice co-founder loop — talk to the Homie out loud, it works for you, takes correction mid-flight, shows its work, and remembers every conversation. Every slice was adversarially gated (Codex two-round + Kimi K3 design review) and live-canary proven.

Highlights

Session memory, both ends of the conversation

  • Session-start identity injection so Talk opens knowing you
  • Session-end vault debrief on both surfaces — dashboard /talk and Discord voice — landing every conversation in the daily log + a searchable episode. The Discord path writes a crash-safe rolling transcript that the lifecycle sweeps in-process (survives the sidecar being killed).

Quick-agent steering

  • Redirect a running background agent by voice: say queues a course-correction that lands at the next turn boundary as a resumed conversation turn (real context, not a restart)
  • cancel stops one cleanly: finish-first compare-and-set + process-tree kill, no orphaned processes
  • New --resume-strict CLI flag for context-safe conversation resumption

Runs visibility

  • A unified run registry with durable JSONL history — every voice-deployed worker survives an API restart and stays visible

Personas / tool-calling

  • Persona capabilities, provisioning, readiness, Discord bindings, blueprints, and one-time capability elevation

Notable fix

  • The Discord voice sidecar's process kill had never actually executed (invalid subprocess.Popen kwargs, silently swallowed) — every /talk leave left a zombie. Fixed and live-proven dead.

Upgrade

Clean pull — no migrations, no config changes required. Mirrors private `3be142ea..bd724223`.

Full history: `git log v1.1.4..v1.2.0`