General-purpose coding policy for Baruch's AI agents
73
91%
Does it follow best practices?
Run evals on this skill
Adds up to 20 points to the overall score
View guide
Low
Low-risk findings worth noting
#!/usr/bin/env bash
# Run one command under a wall-clock budget, for a caller that must finish.
#
# A command that talks to origin (fetch, ls-remote, push, gh) can wait on the
# network for as long as the network likes, and a session start, a stop or a
# round preflight waits with it. `timeout` is GNU coreutils, absent from a
# stock macOS, so the budget is enforced here with python3, which every caller
# already needs.
#
# The budget is a SIGALRM the runner schedules for itself. When it arrives —
# at the budget, or earlier from anyone who sends SIGALRM to the runner's pid
# (how the tests end a budget without waiting on a clock) — the command's
# whole process group gets SIGTERM, then SIGKILL after KILL_GRACE_SEC when
# any member is still running, whether or not the direct child has exited, so
# a git or gh it spawned cannot outlive it. The runner execs python3, so its pid
# is this script's pid.
#
# Contract:
# argv : <seconds> <command> [args...] — seconds is a positive integer.
# stdin : passed to the command.
# stdout: the command's own stdout, unchanged; the runner adds nothing
# (rules/script-delegation.md, the bounded-run.sh carve-out). On
# exit 124 or 125 a caller discards whatever the command printed.
# stderr: the command's own stderr, plus one diagnostic line when the budget
# ran out or the command could not be started.
# exit : the command's own exit code; 124 when the budget ran out (the
# `timeout` convention); 125 on a usage error, a missing python3, or
# a command that could not be started.
# signal: SIGALRM ends the budget now. It is blocked until the command is
# launched and the alarm armed, and blocked again before the alarm
# is cancelled, so an expiry at any point stops the command's whole
# process group and exits 124, never a traceback.
# env : BOUNDED_RUN_TEST_EXPIRE_BEFORE_LAUNCH=1 (tests only) delivers the
# expiry before the command is launched;
# BOUNDED_RUN_TEST_EXPIRE_AFTER_EXIT=1 (tests only) delivers it
# after the command exits, before the alarm is cancelled, and
# =masked delivers it once SIGALRM is blocked again (held pending,
# still exit 124).
set -euo pipefail
#: Seconds between SIGTERM and SIGKILL once the budget is spent.
KILL_GRACE_SEC=2
main() {
if (( $# < 2 )) || [[ ! "$1" =~ ^[1-9][0-9]*$ ]]; then
printf 'bounded-run: usage: bounded-run.sh <seconds> <command> [args...]\n' >&2
return 125
fi
if ! command -v python3 >/dev/null; then
printf 'bounded-run: python3 not found on PATH — install it; %s did not run\n' "$2" >&2
return 125
fi
local program
# The program is an argument, never stdin: stdin belongs to the command.
program="$(cat <<'PY'
import os
import signal
import subprocess
import sys
import time
#: Seconds between checks for a process group still alive inside the grace.
GROUP_POLL_SEC = 0.05
class Expired(Exception):
pass
def expire(signum, frame):
raise Expired()
grace, budget, command = float(sys.argv[1]), int(sys.argv[2]), sys.argv[3:]
ALARM = {signal.SIGALRM}
# SIGALRM stays blocked until the child exists and the alarm is armed: an
# expiry that arrives earlier waits, pending, and is handled below with a
# process group to stop. The child starts with it unblocked again.
signal.pthread_sigmask(signal.SIG_BLOCK, ALARM)
signal.signal(signal.SIGALRM, expire)
if os.environ.get("BOUNDED_RUN_TEST_EXPIRE_BEFORE_LAUNCH") == "1":
# Test seam: an expiry that lands before the command is launched.
os.kill(os.getpid(), signal.SIGALRM)
try:
child = subprocess.Popen(command, start_new_session=True,
preexec_fn=lambda: signal.pthread_sigmask(signal.SIG_UNBLOCK, ALARM))
except OSError as exc:
sys.stderr.write("bounded-run: cannot start {}: {} — check the path and its permissions\n".format(command[0], exc))
sys.exit(125)
def signal_group(sig):
try:
os.killpg(child.pid, sig)
except ProcessLookupError:
pass
def group_alive():
try:
os.killpg(child.pid, 0)
except ProcessLookupError:
return False
except PermissionError:
return True
return True
try:
signal.alarm(budget)
signal.pthread_sigmask(signal.SIG_UNBLOCK, ALARM)
code = child.wait()
if os.environ.get("BOUNDED_RUN_TEST_EXPIRE_AFTER_EXIT") == "1":
# Test seam: an expiry that lands as the command exits.
os.kill(os.getpid(), signal.SIGALRM)
# Cancelled inside the handler's reach: an expiry landing after the wait
# but before this is still stopped here, never a traceback.
signal.pthread_sigmask(signal.SIG_BLOCK, ALARM)
if os.environ.get("BOUNDED_RUN_TEST_EXPIRE_AFTER_EXIT") == "masked":
# Test seam: an expiry that lands once the mask holds it.
os.kill(os.getpid(), signal.SIGALRM)
signal.alarm(0)
if signal.SIGALRM in signal.sigpending():
# Landed after the wait, held by the mask: still an expiry.
raise Expired()
except Expired:
signal.signal(signal.SIGALRM, signal.SIG_IGN)
signal_group(signal.SIGTERM)
# The grace covers the whole group, not the direct child alone: a child
# that exits on SIGTERM can leave a grandchild that ignores it.
end = time.monotonic() + grace
try:
child.wait(timeout=grace)
except subprocess.TimeoutExpired:
pass
while group_alive() and time.monotonic() < end:
time.sleep(GROUP_POLL_SEC)
if group_alive():
signal_group(signal.SIGKILL)
child.wait()
# Only the executable is named: an argument can carry a credential.
sys.stderr.write("bounded-run: {} ran past its {}s budget and was stopped — run it by hand to see where it waits\n".format(
os.path.basename(command[0]), budget))
sys.exit(124)
# A command killed by a signal exits 128 + that signal, as a shell reports it.
sys.exit(code if code >= 0 else 128 - code)
PY
)"
exec python3 -c "$program" "$KILL_GRACE_SEC" "$@"
}
if [[ "${BASH_SOURCE[0]}" == "$0" ]]; then
main "$@"
fi.tessl-plugin
hooks
rules
skills
adopt-fork-pr
herdr-foreman
classify
foreman
references
templates
tests
herdr-standup
migrate-to-plugin
onboard-repo
release
references
tests