SIGN IN SIGN UP

Log a startup timeline for PTY launches

A launch that never binds left no trace of how far it got. A Windows CI
run of real_tool_claude sat ~95s between "starting PTY wrapper" and
Claude's first debug line, and nothing showed which stage held it.

hcom.log now records, per launched agent:
- startup.spawned: hcom and child pids, spawn time, command
- startup.output: the first 8 output chunks with ms since spawn and an
  escaped preview, so terminal setup is told apart from tool output
- startup.dsr_answered (Windows): when the headless reader answered the
  ConPTY cursor query that gates the child's console init
- startup.not_ready: once, if the ready pattern is missing 15s after
  spawn, with chunk/byte counts and when output last arrived

Background runner lines carry timestamps: the PowerShell runner's
"environment ready" and "starting PTY wrapper", and a new "starting PTY
wrapper" line in the Unix script.
A
aannoo committed
36aa4b3709ba69370b9b2a3f8cd228129d6264c5
Parent: 356c6d0