Find the API keys and risky Skills your AI clients can reach
Oriant’s endpoint agent reports which AI provider API keys sit where AI clients can reach them, and it finds and scans every Skill an agent loads. You see the risky Skills first, with the machines that have them.
API keys that AI clients can reach
The device scan reports that an AI provider’s API key is set on a machine and where it is. It reports the name and the location of the key, never its value. The scan sends names, paths and setting values only, and never file contents or secret values. Each report is tied to the person whose email is stamped on the machine, so a reachable key has a person next to it.
Skills installed for AI agents
A Skill is a folder with a SKILL.md: instructions, and sometimes scripts, that an AI client loads on demand. A Skill changes what the agent does and can come from anywhere, such as Anthropic, a marketplace plugin or a teammate’s repo. The agent looks in these places on each machine, and the same Skill in several clones of one repo counts once.
- Claude Code, in ~/.claude/skills
- Claude Code plugins, in the plugin cache
- Codex, in ~/.codex/skills
- The shared agents folder, ~/.agents/skills
- The Claude app, for Skills Anthropic built and Skills a person added
- Repos, in .claude/skills of each local git repo that has a remote
Skill scanning and verdicts
Oriant scans each Skill with the Oriant Skillscanner on its own server, statically and without an LLM. Each version gets a severity and a verdict: Do not install, Caution or Safe. Every finding lists its rule, category, file and line. The same files on many machines are one Skill Version and are scanned once, and an edited Skill is a new version that is scanned again.
Which machines have a risky Skill
The Skills page lists every Skill in the company, worst first. It counts the Skills marked Do not install, Caution and Safe, shows the people with the most flagged Skills, and lists every machine that has a given Skill and which version. Skills marked Do not install also appear under Needs attention on Overview. You can ask your own AI client through Oriant MCP, which has a list_skills tool.
Agent autonomy settings
The device scan also reports how much freedom each agent has on a machine. For Claude Code that means its permission mode, approval policies, hooks, and allow and deny lists. The report holds the setting values, so you can compare machines.
Tools labelled write or destructive
MCP servers label their own tools Read, Write or Destructive, and Oriant shows those labels per connection. MCP Policies use them to decide how far up the scale a team may go. A call to a tool a policy blocks is refused and recorded in the call log. Oriant never labels a tool itself, and a tool with no label is allowed only under the Everything level.
What a Skill scan can and cannot do
Oriant reports a risky Skill. It does not block or remove one, so the Skill has to be removed from those machines separately. A static scan produces false positives, including on some Skills Anthropic ships, so read the finding before you act. A Skill over 10 MiB is listed but not scanned. The agent uploads the files of each Skill once per version, and Oriant deletes a version’s files 30 days after the last machine reported it. The findings stay. Skills scanning runs where the endpoint agent runs, and the self-hosted edition does not include it.
Read the docs
More of Oriant
What can your AI reach today?
Oriant lists the AI clients, MCP servers, API keys and Skills on your machines, and it governs what AI does in your systems.