Adityakk9031/cdebt

Review signals: 1 known vulnerability. 3 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. zod

3 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 →

• cdebt map lists which files hold un-understood AI-written code: Real input produced the expected result: Summary files=108 lines=18148

  Source: Which files hold the most AI-written code no human has verifiably understood?

  Input: cdebt map --no-scip run in the pinned repo

  Expected: a summary table with file counts and AI attribution

  Command: <guest-path> map --no-scip

  Observed: Summary files=108 lines=18148 | AI=100.0% (18148) unknown=0.0% (0) | AI lines verified: 0% | attested (NON-VERIFIED): 0% | AI units verified: 0/953 (95% CI 0%-0%), attested 0 | attribution sources: trailer=18148 * coarse: commit-level trailer heuristic, not line-level.

Examples

• the README 60-second demo builds a throwaway repo and walks map, quiz, map, gate: command completed successfully

  Source: README.md 60-second demo / examples/demo.sh

  Input: bash examples/demo.sh with venv PATH and CDEBT_TRUSTED=1

  Expected: quiz verifies answers and gate reports un-understood lines after edit

  Command: bash examples/demo.sh

  Observed: rstanding expires cdebt gate: FAIL (advisory, exit 0; use --enforce to fail) AI lines in un-understood units: 1 (max 0) | sources: {'trailer': 1} pkg/shop.py::total 1 AI lines NOT UNDERSTOOD (code_changed) [trailer] Some attribution is from commit trailers: coarse, commit-level. cdebt gate: FAIL AI lines in un-understood units: 1 (max 0) | sources: {'trailer': 1} pkg/shop.py::total 1 AI lines NOT UNDERSTOOD (code_changed) [trailer] Some attribution is from commit trailers: coarse, commit-level. (gate exit code 1 with --enforce: expected 1, total() changed and its understanding expired)

Local or hosted: local

Install cdebt from PyPI with the README command and run the README 60-second demo (map, quiz, map, gate), plus cdebt map on the pinned repo

cd <guest-path> && pwd && ls -la && id -u && python3 --version && cat README.md 2>/dev/null | head -c 6000
ls -la /inputs 2>/dev/null; ls -la <guest-path> 2>/dev/null; curl -sS -m 10 "$CDP_URL/json/version"; echo "exit=$?"
cd <guest-path> && sed -n '/## Status/,/## Connect/p' README.md | tail -n +20 | head -c 5000; echo ----; grep -n -i "requires-python\|^name\|^version\|\[project.scripts\]" -A2 pyproject.toml | head -30
cd <guest-path> && grep -n "^## " README.md; echo ----; sed -n '/^## Commands/,/^## [A-Z].*/p' README.md | head -c 5000
cd <guest-path> && sed -n '323,400p' README.md | head -c 3000; echo ----; python3 -m venv --help >/dev/null 2>&1 && echo venv_ok || echo venv_missing; which git node npm uv 2>&1
cd <guest-path> && git init -q && git remote add origin https://github.com/Adityakk9031/cdebt && timeout 120 git fetch --depth 1 origin 406242619b019a458aa2582cc6d0844db1690d67; echo "fetch_exit=$?"; git reset --mixed FETCH_HEAD 2>&1; echo "reset_exit=$?"; git log --oneline -1 2>&1
cd <guest-path> && python3 -m venv <guest-path> && <guest-path> install -q --upgrade pip 2>&1 | tail -3; timeout 600 <guest-path> install --pre "cdebt[agent]" 2>&1 | tail -15; echo "exit=${PIPESTATUS[0]}"
cd <guest-path> && <guest-path> --help 2>&1 | head -60; echo "exit=${PIPESTATUS[0]}"
cd <guest-path> && sed -n '183,200p;226,240p' README.md | head -c 3500
cd <guest-path> && ls examples examples/first_user 2>/dev/null; echo ----; cat examples/demo.sh | head -c 5000
cd <guest-path> && export PATH=<guest-path> && timeout 600 bash examples/demo.sh 2>&1 | tail -c 6000; echo "exit=${PIPESTATUS[0]}"
cd <guest-path> && grep -n -i "trusted\|docker\|sandbox" README.md | head -30; echo ----; sudo -n true 2>&1 && echo sudo_ok; which docker 2>&1
cd <guest-path> && sed -n '301,322p' README.md
cd <guest-path> && export PATH=<guest-path> CDEBT_TRUSTED=1 && timeout 900 bash examples/demo.sh > <guest-path> 2>&1; echo "exit=$?"; sed -n '1,400p' <guest-path> | tail -c 7000
cd <guest-path> && <guest-path> map --no-scip > <guest-path> 2>&1; echo "exit=$?"; tail -c 1500 <guest-path>; echo; git status --short | head; ls -la .cdebt 2>/dev/null | head
cd <guest-path> && grep -rn -i "cron\|schedul\|while True\|--once\|--now\|webhook\|sleep(" src/cdebt/*.py | head -20; echo ----; grep -n "def \(map\|quiz\|gate\)\b" -n src/cdebt/cli.py | head; echo; ls .github/workflows 2>/dev/null; grep -n "schedule" -A3 .github/workflows/*.yml 2>/dev/null | head -10
cd <guest-path> && cat > probe.json <<'EOF'
{"kind":"cli","argv":["<guest-path>","map","--no-scip"],"cwd":"<guest-path>","expected_text":"Summary files=108 lines=18148"}
EOF
cat > claims.json <<'EOF'
[{"kind":"claim","claim":"cdebt map answers which files hold un-understood AI-written code","source":"README.md","source_file":"README.md","quote_source":"README.md","input":"cdebt map --no-scip run in the pinned repo","expected":"a summary table with file counts and AI attribution","probe_kind":"cli","argv":["<guest-path>","map","--no-scip"],"expected_text":"Summary files=108 lines=18148","cwd":"<guest-path>","claim_quote":"Which files hold the most AI-written code no human has verifiably understood?","quote":"Which files hold the most AI-written code no human has verifiably understood?"}]
EOF
python3 -I -c 'import json;json.load(open("probe.json"));print(json.load(open("claims.json"))[0]["quote"])' 2>&1; grep -c "Which files hold the most AI-written code no human has verifiably understood?" ../README.md
cd <guest-path> && cat > claims.json <<'EOF'
[{"kind":"claim","claim":"cdebt map lists which files hold un-understood AI-written code","source":"Which files hold the most AI-written code no human has verifiably understood?","input":"cdebt map --no-scip run in the pinned repo","expected":"a summary table with file counts and AI attribution","probe_kind":"cli","argv":["<guest-path>","map","--no-scip"],"expected_text":"Summary files=108 lines=18148","cwd":"<guest-path>"}]
EOF
cat > examples.json <<'EOF'
[{"kind":"example","claim":"the README 60-second demo builds a throwaway repo and walks map, quiz, map, gate","source":"README.md 60-second demo / examples/demo.sh","input":"bash examples/demo.sh with venv PATH and CDEBT_TRUSTED=1","expected":"quiz verifies answers and gate reports un-understood lines after edit","argv":["bash","examples/demo.sh"],"cwd":"<guest-path>","env":{"PATH":"<guest-path>","CDEBT_TRUSTED":"1"},"expected_text":"2/2 -> UNDERSTOOD"}]
EOF
cat > cycles.json <<'EOF'
{"detected":false,"reason":"Inspected pyproject entry point cdebt=cdebt.cli:app and src/cdebt/*.py: no cron, scheduler, --once/--now, or webhook trigger. Loops found are interactive quiz input (adaptive.py), sandbox wait (sandbox.py), and SCIP handling (scip.py). The pre-push hook is git-triggered and not a product schedule. Commands run once per invocation.","checks":[]}
EOF
for f in claims.json examples.json cycles.json probe.json; do python3 -I -m json.tool $f >/dev/null && echo "$f ok"; done
cd <guest-path> && <env-var>=<guest-path> timeout 900 python3 -I <guest-path> --validate --replay > <guest-path> 2>&1; echo "exit=$?"; tail -c 5000 <guest-path>
cd <guest-path> && python3 -I -c '
import json,sys
t=open("<guest-path>").read()
i=t.find("{")
d=json.loads(t[i:])
print({k:(v if not isinstance(v,(dict,list)) else type(v).__name__) for k,v in d.items()})
for k in ("problems","errors","warnings","valid","ok"):
    if k in d: print(k, json.dumps(d[k])[:1500])
' ; head -c 600 <guest-path>
cd <guest-path> && cat > report.json <<'EOF'
{"outcome":"working","fault_owner":null,"operation":"Install cdebt from PyPI with the README command and run the README 60-second demo (map, quiz, map, gate), plus cdebt map on the pinned repo","operation_scope":"Python 3.12.3 venv at <guest-path>; cdebt 1.1.0rc11 (README requires Python 3.11+). Quiz execution needs Docker or the trusted host sandbox; Docker is not installed here, so CDEBT_TRUSTED=1 was set. Git available. No browser or credentials needed.","steps":["git init; git remote add origin https://github.com/Adityakk9031/cdebt; git fetch --depth 1 origin 406242619b019a458aa2582cc6d0844db1690d67; git reset --mixed FETCH_HEAD","python3 -m venv <guest-path> && <guest-path> install --pre \"cdebt[agent]\"  (README Quick start / 60-second demo)","bash examples/demo.sh (README 60-second demo) with PATH=<guest-path> and CDEBT_TRUSTED=1","<guest-path> map --no-scip on the pinned workspace (README Commands: cdebt map)"],"deviations":["Workspace had no .git; ran git init and the pinned fetch/reset as the task instructed.","Installed cdebt from PyPI (the README's documented command), not from the pinned git source. The PyPI version 1.1.0rc11 matches pyproject.toml.","First demo run failed at the quiz step: 'this repository is not marked trusted, so probes need strong isolation: Docker ... Neither is available' (exit 2). Docker is not installed. Re-ran with CDEBT_TRUSTED=1, the README Safety model's documented host-sandbox switch, on the throwaway demo repo that the demo script generates. The demo's quiz then ran sandboxed with audit+rlimit.","/inputs/coldpass-preflight.py was absent; used <guest-path> as instructed in the task.","cdebt map creates an untracked .cdebt/ directory in the workspace (product output)."],"stuck_step":null,"main_blocker":null,"plain_failure":null,"suggested_fix":null}
EOF
python3 -I -m json.tool report.json >/dev/null && echo report_ok; ls -la <guest-path>; pgrep -fa "cdebt|uvicorn" | grep -v pgrep | head -3; echo done