loadn-ai/loadn

Review signals: 17 known vulnerability. 5 static matches retained as context.

Reviewed as expected for a test run: attempted writes that failed, so nothing was written.

OSV reports a known vulnerability in this dependency version. 11 distinct locations. 17 findings.

5 hardcoded secret matches retained as context; full list in the evidence.

OSV scanned supported lockfiles; unpinned dependencies and unsupported formats are outside this check.

A short run cannot establish that code is safe.

Connections show IP addresses; proxies can hide the final hostname.

Tracing covers dispatched commands and their children, not the whole OS; guest root can interfere with tracing.

Full safety evidence →

• Headless one-shot prompt prints the agent reply to stdout: Real input produced the expected result: LOADN_E2E_OK_7731

  Source: # one-shot, human-readable output

  Input: say hi

  Expected: reply text printed

  Command: <guest-path> -p --dangerously-skip-permissions 'say hi'

  Observed: LOADN_E2E_OK_7731 2026-10-28 13:39:24,415 loadn.tools.grep WARNING rg 不可用——Grep 走纯 Python 扫描兜底（速度慢，rg -g/--multiline 参数语义降级：glob 仍生效、multiline 逐行近似） 2026-10-28 13:39:24,470 loadn.core.models WARNING 模型 'fake' 不在目录（models.json）——窗口按保守 128k 估算（补目录可消除本告警） 2026-10-28 13:39:24,470 loadn.core.cache_warmer INFO cache-warm 不发：cache economics unavailable（目录无价格）

Examples

• Structured JSON result output from the headless quickstart: command completed successfully

  Source: README.md First run section

  Input: say hi

  Expected: JSON result containing the reply text

  Command: <guest-path> -p --output-format json --dangerously-skip-permissions 'say hi'

  Observed: t_tokens": 10000, "cache_creation_input_tokens": 0}, "modelUsage": {"fake": {"inputTokens": 1000, "outputTokens": 200, "cacheReadInputTokens": 10000, "cacheCreationInputTokens": 0, "webSearchRequests": 0, "costUSD": 0.00488}}, "num_turns": 1, "duration_ms": 2495} 2026-10-28 13:39:27,983 loadn.tools.grep WARNING rg 不可用——Grep 走纯 Python 扫描兜底（速度慢，rg -g/--multiline 参数语义降级：glob 仍生效、multiline 逐行近似） 2026-10-28 13:39:28,038 loadn.core.models WARNING 模型 'fake' 不在目录（models.json）——窗口按保守 128k 估算（补目录可消除本告警） 2026-10-28 13:39:28,039 loadn.core.cache_warmer INFO cache-warm 不发：cache economics unavailable（目录无价格）

Local or hosted: local

Install the loadn engine from the pinned source and run a headless one-shot prompt (loadn -p) that prints the reply to stdout

cd <guest-path> && ls -la && cat README.md | head -c 6000 && python3 --version
cd <guest-path> && grep -n -i "quick\|install\|pip \|## " README.md | head -80; ls .coldpass
cd <guest-path> && sed -n 221,312p README.md
python3 --version; which python3.10 python3.11 python3.12 uv 2>/dev/null; echo "CDP=$CDP_URL"; curl -s -m 10 -w "\nHTTP %{http_code}\n" $CDP_URL/json/version
cd <guest-path> && python3 -m venv ~/.venvs/loadn 2>&1 | tail -3; ~/.venvs/loadn/bin/pip install -q -e . 2>&1 | tail -15; echo EXIT=$?; ~/.venvs/loadn/bin/loadn --version 2>&1 | head -3
cd <guest-path> && sed -n 312,350p README.md; grep -rn "LOADN_FAKE_DIR" loadn/*.py loadn/**/*.py | head -8
cd <guest-path> && sed -n 1,140p loadn/providers/fake.py
mkdir -p /tmp/loadn-run/fake /tmp/loadn-run/home && echo "LOADN_E2E_OK_7731" > /tmp/loadn-run/fake/reply && cd /tmp/loadn-run && LOADN_PROVIDER=fake LOADN_FAKE_DIR=/tmp/loadn-run/fake LOADN_HOME=/tmp/loadn-run/home ~/.venvs/loadn/bin/loadn -p --dangerously-skip-permissions "say hi" < /dev/null; echo EXIT=$?
cd <guest-path> && echo $HOME; grep -rn "\-\-once\|\-\-now\|croniter\|def tick\|while True" loadn_webui loadn --include=*.py | grep -v test | head -12; sed -n 160,170p README.md | cut -c1-300
cd <guest-path> && sed -n 115,150p loadn_webui/scheduler.py; grep -n "scheduler\|schedule" loadn_webui/cli.py 2>/dev/null | head; ls loadn_webui | head -50
cd <guest-path> && mkdir -p .coldpass && cat > .coldpass/probe.json <<'EOF'
{"kind":"cli","argv":["<guest-path>","-p","--dangerously-skip-permissions","say hi"],"cwd":"<guest-path>","stdin":"","env":{"LOADN_PROVIDER":"fake","LOADN_FAKE_DIR":"/tmp/loadn-run/fake","LOADN_HOME":"/tmp/loadn-run/home"},"expected_text":"LOADN_E2E_OK_7731"}
EOF
cat > .coldpass/claims.json <<'EOF'
[{"kind":"claim","claim":"Headless one-shot prompt prints the agent's reply to stdout","source":"README.md","source_file":"README.md","quote":"# one-shot, human-readable output","input":"say hi","expected":"reply text printed","probe_kind":"cli","argv":["<guest-path>","-p","--dangerously-skip-permissions","say hi"],"stdin":"","expected_text":"LOADN_E2E_OK_7731","cwd":"<guest-path>","env":{"LOADN_PROVIDER":"fake","LOADN_FAKE_DIR":"/tmp/loadn-run/fake","LOADN_HOME":"/tmp/loadn-run/home"}}]
EOF
cat > .coldpass/examples.json <<'EOF'
[{"kind":"example","claim":"Structured JSON result output from the headless quickstart","source":"README.md First run section","input":"list the todos","expected":"JSON result containing the reply text","argv":["<guest-path>","-p","--output-format","json","--dangerously-skip-permissions","say hi"],"cwd":"<guest-path>","env":{"LOADN_PROVIDER":"fake","LOADN_FAKE_DIR":"/tmp/loadn-run/fake","LOADN_HOME":"/tmp/loadn-run/home"},"stdin":"","expected_text":"LOADN_E2E_OK_7731"}]
EOF
cat > .coldpass/cycles.json <<'EOF'
{"detected":false,"checks":[],"reason":"Inspected README.md and entrypoints. The only scheduler (loadn_webui/scheduler.py, tick()/_loop every CHECK_INTERVAL, exposed via 'loadn-web schedule') belongs to the loadn-web platform, which needs the webui extra, database init and a server; the engine quickstart has no cron/--once/--now trigger. Not exercised."}
EOF
python3 -I -c "import json;[json.load(open('.coldpass/'+f)) for f in ['probe.json','claims.json','examples.json','cycles.json']];print('ok')"
cd <guest-path> && cat > .coldpass/claims.json <<'EOF'
[{"kind":"claim","claim":"Headless one-shot prompt prints the agent reply to stdout","source":"# one-shot, human-readable output","source_file":"README.md","input":"say hi","expected":"reply text printed","probe_kind":"cli","argv":["<guest-path>","-p","--dangerously-skip-permissions","say hi"],"stdin":"","expected_text":"LOADN_E2E_OK_7731","cwd":"<guest-path>","env":{"LOADN_PROVIDER":"fake","LOADN_FAKE_DIR":"/tmp/loadn-run/fake","LOADN_HOME":"/tmp/loadn-run/home"}}]
EOF
sed -i 's/"input":"list the todos"/"input":"say hi"/; s/"source":"README.md First run section"/"source":"README.md First run section"/' .coldpass/examples.json
<env-var>=<guest-path> python3 -I <guest-path> --validate --replay 2>&1 | tail -60
cd <guest-path> && cat > .coldpass/report.json <<'EOF'
{"outcome":"working","fault_owner":null,"operation":"Install the loadn engine from the pinned source and run a headless one-shot prompt (loadn -p) that prints the reply to stdout","operation_scope":"Python 3.12.3 venv at <guest-path>; editable install of the workspace source; LOADN_PROVIDER=fake with a control-file reply (no real model endpoint or API credentials, so no live LLM call was made); no browser needed","steps":["python3 -m venv ~/.venvs/loadn","pip install -e .  (in the venv, from the workspace)","loadn --version -> loadn 0.7.68","LOADN_PROVIDER=fake LOADN_FAKE_DIR=/tmp/loadn-run/fake LOADN_HOME=/tmp/loadn-run/home loadn -p --dangerously-skip-permissions \"say hi\" -> prints LOADN_E2E_OK_7731, exit 0","loadn -p --output-format json (same fake env) -> JSON result, exit 0"],"deviations":["No endpoint or API key available, so the fake provider (documented in README Development setup as LOADN_PROVIDER=fake) was used instead of Anthropic/OpenAI-compatible endpoint; real-model behavior is untested.","Installed from the local workspace with pip install -e . rather than from PyPI or git+https (both are documented in README Install the engine); same package source at the pinned commit.","Full pytest suite (1600+ tests per README) was not run; validation relied on the documented headless CLI operation.","Cycle check: loadn-web scheduler (webui extra) was inspected but not run; detected=false.","Harmless warnings printed to stderr: rg not installed (Grep falls back to pure Python) and model 'fake' not in models.json."],"stuck_step":null,"main_blocker":null,"plain_failure":null,"suggested_fix":"None required for the engine quickstart. Optionally install ripgrep to silence the Grep fallback warning and add a models.json entry for custom model names."}
EOF
python3 -I -c "import json;json.load(open('.coldpass/report.json'));print('ok')"