LeizhenZhang

Vuln Label Dynamic Confirmation Skill — Security skill for Claude Code

Security community

Agent Skill (Claude Code / Codex): dynamic-tool verification of vulnerability labels in BigVul, PrimeVul, MegaVul and DiverseVul.

How to install Vuln Label Dynamic Confirmation Skill

This entry records only its repository, not the path inside it, so there is no exact command to give. Open LeizhenZhang/vuln-label-dynamic-confirmation-skill and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Vuln Label Dynamic Confirmation Skill does

Agent Skill (Claude Code / Codex): dynamic-tool verification of vulnerability labels in BigVul, PrimeVul, MegaVul and DiverseVul.

Alternatives in Security

Skills — Trail of Bits Claude Code skills for security research, vulnerability detection, and audit workflows 4k ★
Clawsec — Security skill suite with drift detection, automated audits, and skill integrity verification 798 ★
Raptor Loop Hunt — RAPTOR autonomous looping multi-altitude security vulnerability hunt — Claude Code skill 212 ★

README

vuln-label-dynamic-confirmation

An [Agent Skill](https://agentskills.io) that turns a long research operating manual into an installable, tool-agnostic skill. It tells a coding agent how to verify vulnerability labels in the BigVul, PrimeVul, MegaVul and DiverseVul datasets with dynamic tools (sanitizers, two-sided differentials, behavioural oracles), how to decide label noise honestly, what evidence must be left on disk so that anyone can re-run it, and how to sign off.

The same folder works for **Claude Code** and for **OpenAI Codex** (the coding agent in ChatGPT, the Codex CLI and the Codex desktop app): both read the Agent Skills format, a folder with a `SKILL.md` carrying YAML front matter plus optional `references/` and `scripts/`.

What is in the folder

vuln-label-dynamic-confirmation/
├── SKILL.md                            the skill body: operating order, hard rules, per-dataset parameter table
├── README.md                           this file
├── references/
│   ├── BIGVUL_LOOP_PROMPT.md           full English translations of the four long-form prompts
│   ├── PRIMEVUL_LOOP_PROMPT.md
│   ├── MEGAVUL_LOOP_PROMPT.md
│   ├── DIVERSEVUL_LOOP_PROMPT.md
│   ├── PROJECT_AGENT_BRIEF.md          per-project working brief
│   ├── CONFIRMATION_SCOPE.md           note on confirmation scope (defect_site vs reachability)
│   ├── primevul_outcome_taxonomy.py    the four outcomes and their secondary-field vocabularies
│   └── SOURCE.json                     origin path and sha256 of every reference
└── scripts/
    ├── run_checkers.sh   [ids]   sign-off checkers
    ├── refresh_stats.sh                   statistics and report refresh
    └── sync_references.sh                          re-sync references/ from the repository originals

`SKILL.md` is the compressed version. The long-form prompts in `references/` are the final authority on criteria; when they disagree, the dataset's own prompt wins, and among the four prompts the Bi