jason.woltjeandClaude Opus 5.5
771fc3d270
Pi 0.85.1 and Claude Code 2.1.289 (Agent SDK 0.3.289), 34 cases against a scripted local model in a loopback-only network namespace. MATRIX.md has the command, outcome, model-visible message and fail-closed column per case, and one rely-on line per case. Evidence under evidence/. Co-Authored-By: Claude Opus 5.5 <[email protected]>
24 lines
1.1 KiB
Bash
Executable File
24 lines
1.1 KiB
Bash
Executable File
#!/usr/bin/env bash
|
|
# Row S0: runs one probe inside the network namespace that probe.sh sets up.
|
|
# Starts the mock model on loopback, runs the harness under a wall-clock
|
|
# limit, and writes outcome.txt. Usage: inner.sh <run-dir> <limit-s> -- cmd...
|
|
set -u
|
|
run=$1 limit=$2
|
|
shift 3
|
|
here=$(dirname "$(readlink -f "$0")")
|
|
cd "$run/work" || exit 90
|
|
node "$here/mock-anthropic.mjs" 2> "$run/mock-stderr.txt" &
|
|
mock=$!
|
|
for _ in $(seq 100); do grep -q listening "$MOCK_LOG" 2>/dev/null && break; read -rt 0.05 <> <(:) || true; done
|
|
start=$(date +%s%N)
|
|
timeout -k 5 "$limit" "$@" < /dev/null > "$run/stdout.txt" 2> "$run/stderr.txt"
|
|
rc=$?
|
|
end=$(date +%s%N)
|
|
kill "$mock" 2> /dev/null
|
|
{
|
|
echo "uid=$(id -u) net=$(ip -o link | awk -F': ' '{print $2}' | tr '\n' ' ')"
|
|
echo "exit=$rc elapsed_ms=$(((end - start) / 1000000))"
|
|
if [ -e "$run/work/target.txt" ]; then echo "target=present content=$(head -c 80 "$run/work/target.txt" | tr '\n' ' ')"; else echo "target=absent"; fi
|
|
echo "gate_loaded=$(cat "$GATE_LOG" 2> /dev/null | grep -c ' loaded ') gate_calls=$(cat "$GATE_LOG" 2> /dev/null | grep -cE ' invoked | input=')"
|
|
} > "$run/outcome.txt"
|