# Valor

An autonomous AI coworker that receives tasks over Telegram or email, plans and critiques its own work, writes code, runs tests and opens pull requests, restarting or hibernating itself on failures.

Valor wraps an existing agent harness, such as Claude Code, and bridges it to the communication channels its supervisor already uses, including Telegram, email and LinkedIn, so tasks are assigned and replied to in the same thread rather than through a separate dashboard. A subconscious memory system injects relevant past decisions and corrections into a session at the right moment, and a watchdog and self-healing pipeline restarts the worker on a crash, hibernates it on repeated API failures, and escalates to a person only when it cannot recover on its own.

Its software development pipeline moves work through plan, critique, build, test, patch and documentation stages, assigning each stage to the model suited to its cognitive load, and treats the pipeline as a graph with failure cycles rather than a straight line, so a critique can loop back to planning when a concern surfaces. New capabilities are added by dropping a skill file into a skills directory, wiring a tool, or connecting a new communication channel.

Website: https://github.com/tomcounsell/ai