)]}'
{
  "log": [
    {
      "commit": "6fe809c2282f5d7e258ab60f4ff4d0b3e0739332",
      "tree": "1f9a0bccf5ab23ff15eb38c26bda8c9d4e3dfb3d",
      "parents": [
        "4a370aa1fe7a2fcafc41c0cd1bb4bf59bfb070a2"
      ],
      "author": {
        "name": "Josh Bleecher Snyder",
        "email": "josharian@gmail.com",
        "time": "Thu Jul 24 16:22:51 2025 -0700"
      },
      "committer": {
        "name": "Josh Bleecher Snyder",
        "email": "josharian@gmail.com",
        "time": "Wed Jul 30 11:30:10 2025 -0700"
      },
      "message": "claude: bash tool smorgasbord\n\nThis is a combination of a bunch of changes and bug fixes\naccumulated over a week. It decided not to spend a bunch of time\nteasing it apart into its components. I apologize.\n\nSignificant changes include:\n\n- clean up background bash execution\n- improve documentation\n- remove TODO we\u0027re not gonna do\n- add a \"process completed\" note to stderr\n- convert background result printing to xml-ish\n- combine stdout and stderr in background bash to match foreground use\n- hint to sketch to kill the process group\n- add missing timeouts propagation\n- tell agent that bash calls are stateless\n- thread pwd through explicitly\n- unify command creation\n- speed up missing command installation\n    I tried a bunch of different ways to prompt engineer this to be faster.\n    But Claude will be Claude. Solution: switch from agent to one-shot.\n\n    This is marginally more brittle, and can only use a package manager,\n    but that also prevents a bunch of possible curl|bash mess, etc.\n    And this was always best-effort anyway.\n\n    It\u0027s now MUCH faster to fail on non-existent commands, and about 2x\n    faster on the real commands I tried (yamllint, htop)...now mostly down\n    to the irreducible work involved in actually doing the installation.\n- remove SKETCH_ from bash env, except SKETCH_PROXY_ID\n- delay kill instructions until actually needed\n- add simple GIT_SEQUENCE_EDITOR\n- overhaul cancellation\n- explicitly disable EDITOR to prevent hangs\n    I have big plans here, but this will do for now.\n- simplify and unify handling of long outputs\n- switch to center trunctation of long outputs\n- add zombie process cleanup using unix.Wait4\n    Wow, I tried a bunch of things here.\n\n    When running as PID 1, we are responsible for reaping zombies.\n    Unfortunately, we can\u0027t do this in the simple/obvious way,\n    because simply listening for SIGCHILD and reaping races\n    with running cmd.Wait. We can\u0027t use a separate init process\n    or double-init sketch, because then we lose our seccomp\n    protection, and there\u0027s no particularly good way to extend it.\n\n    Instead, (h/t to Philip asking a good question), observe\n    that we are in a very controlled environment, and pretty much\n    the only way to get zombies is via the bash tool.\n    So we add reaping tied specifically to process groups started\n    by the bash tool, with an explicit understanding of their lifecycle.\n\n    Auto-installation of tools still creates zombies.\n    We now know how to fix it, but it is rare, so who cares.\n"
    }
  ]
}