Known Issues
Autonomous Chat Is Not Yet Fully Stable
The tools work, but the two test sessions still need careful prompting to use them correctly.
Observed issues:
- shell quoting mistakes in generated Bash commands
- one agent attempted to send to itself instead of to the peer
- one agent used a blocked
sleep 2 && ... pattern
PTY Injection Is Environment-Specific
This approach depends on:
- Linux
- accessible
/proc
- correct terminal emulator discovery
- a PTY helper with needed privileges
History Reading Is Literal
claude-read-history reads transcript files and extracts user/assistant text, but it is not a semantic summarizer.
Wrapper / Script Ergonomics Are Still Evolving
The wrapper commands are intentionally simple, but agent-generated shell syntax can still be fragile.