vstack-s5-redteam — Claude Code skill for model attacks
The vstack-s5-redteam Claude Code skill probes a model and its gates with independent attacks. It runs at stage S5 to expose weak checks and verify findings against real code.
MIT ★ 89 updated 2026-11-16
/plugin marketplace add hao-ai-lab/vstack
/plugin install vstack@hao-ai-lab-vstackWhat vstack-s5-redteam does
The skill tests model fidelity, property and gate strength, and claims that extend to a larger configuration. It takes the complete S4 bundle and its scoped statement with hashes as input. Each finding records the command, observed and intended output, source anchors, and confidence.
The coordinator reruns findings and groups shared causes. Every independently confirmed failure stays on the list. Where feasible, a successful bypass becomes a mechanical check, followed by the matching regression attack after gate changes. The red-team role edits only its own artifact copies and leaves reviewed originals untouched. A complete generic attack suite is not bundled.
What it helps with
- detect mismatches between a model and real behavior
- probe properties and gates for bypasses
- test model claims against a larger configuration
- record confirmed failures for coordinator review
Example prompts for the agent
- Run stage S5 against the complete S4 bundle and record observed and intended output for each attack.
- Probe the gates for bypasses and rerun the matching attacks after each change.
- Send confirmed findings to the coordinator without changing reviewed originals.
How to install vstack-s5-redteam
1Plugin marketplace
/plugin marketplace add hao-ai-lab/vstack/plugin install vstack@hao-ai-lab-vstackThis installs the whole vstack plugin with all its skills
2Skills CLI (Vercel)
npx skills add hao-ai-lab/vstack --skill vstack-s5-redteam -a claude-code3Manually: put the files in place
~/.claude/skills/vstack-s5-redteam/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/vstack-s5-redteam/
1Skills CLI (Vercel)
npx skills add hao-ai-lab/vstack --skill vstack-s5-redteam -a codex2Manually: put the files in place
.agents/skills/vstack-s5-redteam/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add hao-ai-lab/vstack --skill vstack-s5-redteam -a cursor2Manually: put the files in place
.cursor/skills/vstack-s5-redteam/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install hao-ai-lab/vstack skills/vstack-s5-redteam2Manually: put the files in place
.github/skills/vstack-s5-redteam/SKILL.md1Skills CLI (Vercel)
npx skills add hao-ai-lab/vstack --skill vstack-s5-redteam -a gemini-cli2Platform CLI
gemini skills install https://github.com/hao-ai-lab/vstack --path skills/vstack-s5-redteamManually: put the files in place
.devin/skills/vstack-s5-redteam/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about vstack-s5-redteam
How do I install vstack-s5-redteam in Claude Code?
Add the marketplace with /plugin marketplace add hao-ai-lab/vstack, then install the skill with /plugin install vstack@hao-ai-lab-vstack.
What does vstack-s5-redteam do?
It attacks a model and its checks, records evidence for findings, and sends them to a coordinator for independent reruns.
What does vstack-s5-redteam need to run?
It needs the complete S4 bundle and its scope statement with hashes.
Source text
SKILL.md · MIT
--- name: vstack-s5-redteam description: "Audit calibrated verification through independent attacks on fidelity, property/gate strength and scope extrapolation." --- # S5 — Attack model fidelity and gate strength Read [Layer 0: shared operating rules](../../docs/layer-0.md), this stage and the assigned input bundle. ## Role and write scope Red team reads reviewed artifacts and mutates only its own copies. It reports findings; the coordinator verifies them and assigns fixes. ## 1. Input The complete S4 bundle and its scope statement, with hashes. ## 2. Procedure 1. When delegation is available and authorized, assign independent attacks: test fidelity with missing or extra behaviors; test properties and gates with bad models, weak specs and forged output; test scale with one larger configuration. Attack the [open-environment rules](../vstack-s3-model/SKILL.md#open-environment): a stimulus whose `InputEnabled_<E>` is missing or vacuous, a guard that is neither a code condition nor a registered bound, a pruned failure path, fairness on an environment action, an `EnvContract_*` without a documentation quote, and a named restriction inside a passing configuration. 2. For each finding record exact command, actual output, intended output, source anchors and confidence. Mark findings as execution-supported when a run supports them, or as reproduced on real code. Items without run evidence remain leads. 3. The coordinator reruns findings and merges common causes. Keep every independently confirmed failure, not just one reviewer's list. 4. Turn a successful bypass into a mechanical check where feasible and record its ablation. Rerun the corresponding attack after gate changes. 5. On the full adequacy track, confirm each seeded mutant and each revert from the blind set on real code before it counts, then hand scoring to a role that did not build the model ([scoring](../../docs/model-adequacy.md#scoring)). ## 3. Exit checks - Resolve each execution-supported or reproduced finding, or file a rollback ticket. The coordinator reviews these manually. - Update the attack-to-enforcer table with evidence and remaining gaps. - Run the corresponding attack regressions for every gate change. The source has an ASSUMES selftest; a complete generic attack suite is not bundled. - S5 is mandatory; a model author's success report is insufficient. - Apply [Gate 2](../../docs/exposure-gate.md#gate-2-red-team) to every finding S6 hands back: record the refutation attempt (outer guard, documented intent, a freedom the real system lacks) and whether it failed. A finding without this record cannot be listed as verified. ## 4. Rollback Return fidelity defects to S3 and a weak ideal to S2 with property-owner approval. Assign weak verdicts or gates to the coordinator for repair. Never modify the reviewed originals from the red-team role. Use [a rollback ticket](../../templates/rollback.md). ## Concrete Examples For PD KV slot ownership, attack a copied ideal action, an unchecked mapping assumption and an implementation that silently discards a posted write. These exercise mirror, ASSUMES and specification-mutation checks.
© Hao AI Lab · MIT · hao-ai-lab/vstack