Tutorial: grade LLM Honeypot (Palisade) with UHBS¶
Status: Informative · evaluation proof
Target: https://github.com/PalisadeResearch/llm-honeypot · commit 156004a1b122f201448635417ee47bd44d7f28ca
Protocol graded: SSH :2222 (SFTP subsystem on). Telnet is disabled in the shipped configs/cowrie.cfg.
0. Prerequisites¶
git clone https://github.com/uhbs/uhbs-standard.git
cd uhbs-standard
pip install -c constraints.txt -e ".[dev,lab]"
docker pull cowrie/cowrie:latest
docker network create uhbs-lab 2>/dev/null || true
1. Clone source (Module F)¶
mkdir -p .local/labs
git clone https://github.com/PalisadeResearch/llm-honeypot.git .local/labs/llm-honeypot
cd .local/labs/llm-honeypot
git checkout 156004a1b122f201448635417ee47bd44d7f28ca
Lab overlays (copied under UHBS): ../../labs/llm-honeypot/configs/.
2. Start Cowrie with LLM-honeypot overlays¶
Do not replace src/cowrie/commands/__init__.py from this fork on Cowrie ≥3.x — it breaks the Twisted cowrie plugin. Mount individual trap modules instead:
LABCFG="$PWD/docs/conformance/labs/llm-honeypot/configs"
docker rm -f llm-honeypot-lab 2>/dev/null || true
docker run -d --name llm-honeypot-lab --network uhbs-lab \
-p 127.0.0.1:12222:2222 \
-v "$LABCFG/cowrie.cfg:/cowrie/cowrie-git/etc/cowrie.cfg:ro" \
-v "$LABCFG/userdb.txt:/cowrie/cowrie-git/etc/userdb.txt:ro" \
-v "$LABCFG/motd:/cowrie/cowrie-git/honeyfs/etc/motd:ro" \
-v "$LABCFG/version:/cowrie/cowrie-git/honeyfs/proc/version:ro" \
-v "$LABCFG/passwd:/cowrie/cowrie-git/honeyfs/etc/passwd:ro" \
-v "$LABCFG/whoami.py:/cowrie/cowrie-git/src/cowrie/commands/whoami.py:ro" \
-v "$LABCFG/ps.py:/cowrie/cowrie-git/src/cowrie/commands/ps.py:ro" \
-v "$LABCFG/df.py:/cowrie/cowrie-git/src/cowrie/commands/df.py:ro" \
-v "$LABCFG/pwd_cmd.py:/cowrie/cowrie-git/src/cowrie/commands/pwd_cmd.py:ro" \
-v "$LABCFG/ls.py:/cowrie/cowrie-git/src/cowrie/commands/ls.py:ro" \
-v "$LABCFG/llm_prompts.py:/cowrie/cowrie-git/src/cowrie/commands/llm_prompts.py:ro" \
-v "$LABCFG/cat8193.py:/cowrie/cowrie-git/src/cowrie/commands/cat8193.py:ro" \
cowrie/cowrie:latest
until docker logs llm-honeypot-lab 2>&1 | grep -q 'Ready to accept SSH'; do sleep 1; done
# Credentials: root / admin (see labs/.../userdb.txt)
3. SSH quick + full¶
mkdir -p docs/conformance/reports/llm-honeypot/ssh/{quick,full}
UHBS_QUICK=1 UHBS_AIRGAP_ATTESTED=1 \
uhbs-lab \
--inventory .local/llm-honeypot-inventory.yaml \
--target llm-honeypot-ssh \
--tps docs/conformance/labs/llm-honeypot/low_interaction_ssh_quick.yaml \
--protocol ssh \
--phases profile,static,sandbox,dynamic,score --modules A,B,C,D,E,F \
--quick --skip-sast-tools --concurrency 10 --requests 50 \
--out docs/conformance/reports/llm-honeypot/ssh/quick \
--environment "Quick Docker lab: llm-honeypot-ssh"
UHBS_AIRGAP_ATTESTED=1 \
uhbs-lab \
--inventory .local/llm-honeypot-inventory.yaml \
--target llm-honeypot-ssh \
--tps docs/conformance/labs/llm-honeypot/low_interaction_ssh_full.yaml \
--protocol ssh \
--phases profile,static,sandbox,dynamic,score --modules A,B,C,D,E,F \
--concurrency 25 --requests 200 \
--out docs/conformance/reports/llm-honeypot/ssh/full \
--environment "Full Docker lab: llm-honeypot-ssh"
Local inventory maps host 127.0.0.1:12222 → container :2222 (avoids clash with other labs on :2222).
What you get from this lab¶
This tutorial reproduces the published UHBS evaluation proof for analysts who need to verify numbers, not trust a screenshot. After a successful run you should have:
SCORECARD.txt— verbatim module table, UHQS, letter grade, and δ_C Safety Gatereport.json— machine-readable scores for automation / diffing- Optional harness logs under the lab telemetry directory
How CTI / blue team should use the artifacts¶
- Open the full SCORECARD first (authoritative). Treat quick as a smoke check unless the methodology says otherwise.
- Read modules A–F with READING-UHQS.md: low B is often “credential sink by design,” not a broken decoy.
- Confirm δ_C = 1.0 (or understand why containment failed) before citing UHQS externally.
- Wire your own log shipping; UHBS Module C is harness visibility, not SIEM coverage.
Trust limits¶
- UHBS 4.2.2 evaluation proof is informative — not a certification, endorsement, or ranking.
- Product names appear only under
docs/conformance/as evaluation evidence. - Re-run after upstream or TPS changes; do not invent scores without regenerating artifacts.