codex
Verdict safe, but 2 findings (2 medium), mainly social engineering, unauthorized tool use. Most recent scan 2026-02-05.
This skill has MEDIUM security concerns primarily related to deceptive technical claims and unrestricted external command execution. The skill claims to use non-existent GPT-5.2 models and references deprecated OpenAI Codex, which constitutes social engineering through false capabilities. The core functionality executes an external 'codex' CLI tool with various sandbox modes including 'danger-full-access' that permits broad system access and network operations. The use of stdin piping for command execution and systematic stderr suppression could mask security issues. While no active data exfiltration or credential theft was detected, the combination of fictional model claims, unrestricted command execution, and missing security metadata creates moderate risk. The skill would benefit from accurate technical claims, input validation, restricted default permissions, and complete metadata disclosure.
Scanner's overall assessment (LLM analyzer)
Second scanner: NVIDIA SkillSpector
The two scanners disagree: Cisco found nothing and SkillSpector says do not install. One of them has a blind spot here; read the skill before trusting either. How the two scanners compare across the whole corpus: Two scanners, one corpus.
What this skill does
Use when the user asks to run Codex CLI (codex exec, codex resume) or references OpenAI Codex for code analysis, refactoring, or automated editing. Uses GPT-5.2 by default for state-of-the-art software engineering.
- Package
- 2 files, 7.2 KB
- SKILL.md
- 721 words, 7 sections
- Code
- None. Instructions only.
SKILL.md covers: Codex Skill Guide · Running a Task · Quick Reference · Model Options · Following Up · Error Handling · CLI Version
Findings
MEDIUM Deceptive Model Claims - Non-existent GPT-5.2 Models Social Engineeringllm
The skill claims to use 'GPT-5.2', 'gpt-5.2-max', 'gpt-5.2-mini', and 'gpt-5.1-thinking' models which do not exist in OpenAI's current model lineup. As of 2024, OpenAI's latest models are GPT-4 series. This appears to be either fictional content or an attempt to mislead users about the capabilities being offered. The skill references 'OpenAI Codex' which was deprecated in March 2023. This constitutes social engineering through false technical claims.
SKILL.md
Model claims: 'gpt-5.2' (flagship), 'gpt-5.2-max', 'gpt-5.2-mini', 'gpt-5.1-thinking' with specific pricing ($1.25/$10.00) and performance metrics (76.3% SWE-bench). References deprecated 'OpenAI Codex' in description.MEDIUM Unrestricted Command Execution via External CLI Tool Unauthorized Tool Usellm
The skill executes an external 'codex' CLI tool with user-controlled prompts and various sandbox modes including 'danger-full-access' which permits network access and broad system permissions. The skill uses stdin piping for resume functionality ('echo "prompt" | codex exec') which could be exploited for command injection if the external tool doesn't properly sanitize input. The --full-auto flag enables autonomous operation without confirmation, and --skip-git-repo-check bypasses safety checks.
SKILL.md
Commands like: 'codex exec --sandbox danger-full-access --full-auto --skip-git-repo-check' and 'echo "your prompt here" | codex exec --skip-git-repo-check resume --last 2>/dev/null'. The danger-full-access mode explicitly permits network and broad system access.What is in the package
| File | Kind | Size |
|---|---|---|
README.md | doc | 2.1 KB |
SKILL.md | doc | 5.1 KB |
External hosts referenced
github.com
Hostnames that appear in the skill's files. Referencing a host is not a finding; compare against the findings above and the skill's stated purpose.
Package contents recorded 2026-10-22 at commit b00cf8491de1. The scan above ran against the version current at scan time; the two can differ if the skill has since been updated.
What we checked
- Static analyzer. Pattern rules over SKILL.md and bundled files: known-bad commands, exfiltration URLs, secrets, obfuscation.
- Behavioral analyzer. Checks what the skill asks the agent to do against what its description claims it does.
- Llm analyzer. A language model reads the skill and reports intent-level issues the rules cannot express.
- Trigger analyzer. Looks for instructions that fire on conditions the user did not ask for.
- Meta analyzer. A second model pass that validates, deduplicates and re-scores the other analyzers' findings.
What we did not check
- Runtime behaviour. The skill was not executed; findings come from reading its files.
- Anything fetched at install or run time, including remote scripts, package dependencies, and URLs the skill tells the agent to open.
- Content the analyzers cannot read: images, binaries, and some non-English text.
- Known scanner bypass classes. See the methodology and limitations page.
Sources
- Repository: https://github.com/softaworks/agent-toolkit
- Raw scan result: codex-scan.json
- Badge for your README:
- Request a re-scan