Semia CLI
Semia is a skill-auditing CLI that reads instructions as data without executing the commands in the skill under review.
- 615 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- Analysis sends text to the selected model, writes reports and can prepare an instruction patch.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/semiaDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Analysis sends text to the selected model, writes reports and can prepare an instruction patch.
This is third-party code. Review the repository files before installing.
What it does
Semia is a skill-auditing CLI that examines instructions before an agent loads them. It prepares the source, asks the selected model to extract possible actions, evaluates those actions against rules and produces a report. Findings point to source lines so users can see the basis for each conclusion. Reports are available as Markdown, JSON or SARIF for GitHub. After analysis, `semia repair` creates a revised `SKILL.md` from the findings for users to review before replacing the instruction.
Who it is for. Reviewers examining third-party skill commands and data access before enabling an instruction in an agent.
Good fit when
- Before enabling a skill, a reviewer examines its documented commands.
- An instruction is checked for secret reads and network actions.
- Audit results are sent to GitHub Code Scanning.
- After reviewing findings, an author prepares an instruction patch.
Not a fit when
- You require proof that all harmful behavior is absent.
- You expect sandboxed execution of the skill.
- The instruction text cannot be sent to the selected model.
Example request
Audit this skill with Semia without running it and show commands, secret reads and source lines supporting the findings.Limitations
The CLI needs Python 3.11 or later and a configured provider for behavior extraction. Instruction text is sent to the selected model, so use an authorized API or your installed CLI. Static analysis can miss an action or misunderstand a condition; inspect important findings against source lines. The repair command creates a patched SKILL.md that needs ordinary diff review before it replaces the working instruction.
How to disable. Uninstall semia-audit and disable any installed plugin. Reports and proposed patches remain under .semia in the project.
FAQ
Is the audited skill executed?
No. It reads the instruction as data and uses a model to extract possible actions. The result does not establish how the skill will behave during actual execution.
Where are results written?
Under `.semia/runs/` by default; change it with `--out`.
Can I work without a separate API key?
A plugin uses its host model, and installed Claude Code or Codex CLIs are also supported.
How do I obtain SARIF?
Use `semia report` to export a scan; `--format sarif` selects SARIF. The example is `semia report .semia/runs/some-skill --format sarif`; replace some-skill with your run directory and submit the report to GitHub Code Scanning.
Which stages are deterministic?
Preparation, rule evaluation and reporting; behavior extraction uses a model.
Discussion
No questions or reviews yet.
Related
CLI toolsThree Semgrep skills for Claude Code and Cursor: secure coding by OWASP, LLM application security, and running Semgrep with custom rules
A security scanner for agent skills: finds prompt injection, data exfiltration and malicious code before you install
Security scanner for agent skills and MCP servers: finds prompt injection, data exfiltration and supply chain risks before install
Static code analysis with rules that look like source code, plus a built-in MCP server for AI agents
Cybersecurity Skills
Cybersecurity Skills for AI Agents
42 security skills for Claude Code, Cursor and Codex: code and cloud audits, authorization-gated pentesting and incident response
VibeSec
VibeSec-Skill
A secure coding skill: the agent looks at code from a bug hunter's perspective and prevents common web vulnerabilities
README in short
Inside a host plugin, the host model extracts facts while a bundled utility prepares data and evaluates rules.