mirror of
https://github.com/usestrix/strix.git
synced 2026-10-14 04:08:06 +00:00
b1174ed95a
A user message already resumes a failed agent's lifecycle from where it stopped. `send` lifts that agent's human-wake gate, `wait_for_message` returns, the loop resets the nudge and idle-resume budgets - "real input is real progress" - and `mark_running` retires the error before the next turn. Any agent can be addressed: `_send_message` takes an arbitrary agent id, so a failed child is reachable by selecting it, exactly as a parked one is. So there was nothing here to fix. The premise behind these commits - that only the root is reachable, stranding every other agent that dies - was wrong, and each revision built a larger mechanism on it: a scan-wide revival, then an attempt budget to bound the runaway that revival introduced. What remains is the TUI, which is where the reported symptom always was: an error that had already resolved stayed on the screen. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> |
||
|---|---|---|
| .. | ||
| agents | ||
| config | ||
| core | ||
| interface | ||
| llm | ||
| report | ||
| runtime | ||
| skills | ||
| telemetry | ||
| tools | ||
| utils | ||
| __init__.py | ||