Before trust
Supply Chain is the part of Signals Corps that finds every AI agent installed on your teams' devices, checks every skill those agents load, and controls which components may run. Know which agents every team runs, from a recruiter's resume skill to an engineer's coding agent, including the ones IT never approved, and block a risky skill on every device. The Signals Corps Endpoint Agent audits every device for desktop assistants such as the Claude and ChatGPT desktop apps, coding agents such as Claude Code, Codex, Cursor, and Windsurf, and the skills, plugins, connectors, and MCP servers they can load, then checks every skill on the device.
Free devices on every plan, then priced per device. See pricing
device:endpoint_4c19 trust:any⌘↵New agent found · Unreviewed
Critical finding · Blocked · Applied
Identified · Trusted
Every device, every agent
Recruiters add resume skills, sales teams add CRM connectors, and engineers add coding agents in minutes, often without telling IT.
The Signals Corps Endpoint Agent on each device audits them all: which agents are installed, which versions, and everything each one can load.
Know every agent your teams run, including the ones IT never approved. Supply Chain finds the Claude and ChatGPT desktop apps, Claude Code, Codex, Cursor, Windsurf, and other agents and IDE extensions on each device, with their versions and the skills, plugins, connectors, MCP servers, and hooks each one can load.
Find prompt injection, credential theft, data exfiltration, and unsafe execution in skills before your teams trust them. Every installed skill is checked on the device with Supply Chain Agent Threat Rules (ATR), and the skill's content stays on the device.
Stop an untrusted skill or plugin on every device, and know the block is in force. Administrators mark components trusted or blocked, a blocked skill or plugin stops running in the agent, and Supply Chain shows whether each block is applied on each device.
How it works
The Signals Corps Endpoint Agent audits each device on a regular schedule for installed agents, their versions, their components, and the identities they can reach.
The Endpoint Agent checks each skill on the device and reports findings, content hashes, and provenance. The skill itself is never uploaded.
Administrators trust or block each component. Unknown components are monitored until someone decides.
Blocks reach each agent through its managed settings and hooks, and each device reports whether the block is applied.
Built for rollout
Roll out in one step by installing the Signals Corps Endpoint Agent with your MDM. Runtime runs in the same Endpoint Agent, so one install covers them together.
Prompts, skill content, file paths, device usernames, and hostnames stay on the device. Devices report component identity, hashes, findings, coverage, and the account names of identities agents can reach, encrypted so only administrators can reveal them.
Employees in every team see the inventory and findings, so they know what is trusted. Owners and administrators enroll devices and decide what to trust or block.
Handle Supply Chain work without leaving your agent. The Signals Corps MCP server lets agents read inventory and findings, and trust or block a component once you confirm.
severity:critical rule:ATR-2026-01988SKILL.md lines 41 to 47
Personal skill · sha256:9f2c…41ab
Claude Code 8 · Codex 3 · Cursor 1
Applied on 11 · New session needed on 1
Evidence people can act on
Administrators can make the call, and IT & Security teams can defend it, because every finding identifies the device, the agent, the exact skill version, the matched rule, its severity, and the recommended action.
Identity discovery
Know what an agent could reach if it misbehaves: the Endpoint Agent maps the identities each agent is configured to use or may inherit.
See each agent's reach across code hosts, clouds, clusters, and servers. The Endpoint Agent inspects agent configuration and developer identity sources such as GitHub CLI accounts, cloud profiles, Kubernetes contexts, and SSH agent metadata.
Judge each link by its evidence. Select a device or agent in Control Plane to see its relationships, for example that Codex is configured with a GitHub identity through an OAuth session. Each relationship says whether the agent is configured with the identity, references a credential source, or has local evidence that the source exists.
Map identity risk without exposing a secret. The Endpoint Agent reads account names and non-secret metadata only, and never reads passwords, tokens, private keys, browser sessions, or vault contents.
ChatGPT desktop app · HRWorkday connectorConfigured
Claude desktop app · SalesSalesforce sign-inMay inherit
Cursor · PlatformToken in MCP configFinding
Codex · EngineeringGitHub accountConfigured
Supply Chain capabilities
Track every AI agent and version across enrolled devices, including the ones IT never approved.
New components start as monitored, so rollout adds visibility before friction. Switch to a stricter default when you are ready.
Keep approvals honest when skills change. Trust a specific version or a reviewed publisher, and a skill whose content changes needs its own decision.
Take a risky skill out of use on every device by stopping it in the agent or disabling the plugin that provides it.
Know where each block is in force. Supply Chain shows which devices need a new session to apply it, since a skill already loaded stays until that session ends, and which versions are not supported.
Report findings in the frameworks your analysts and auditors already use, with each one mapped to MITRE ATLAS techniques and the OWASP Top 10 lists for LLM and agentic applications.
An MCP server can now serve Agent Skills as well as tools. This post covers what that changes for the agents your staff run, which clients load them today, and what to check now....
Team Sales Engineering
Signals Corps Control Plane
Set policy, see coverage, stop high-risk actions, and investigate incidents across every agent and team from one operational view.
Write a rule once and have each control point apply it.
Know who is behind every agent, with users, devices, teams, and tools connected.
Defend every decision, with the rule, relevant details, and outcome kept together.
Roll out with one endpoint agent, then manage exceptions, approvals, and reporting in one place.
While agents act
Stop risky uploads, sends, and commands in real time while safe work in every team carries on.
Early access
Start reducing agent risk before general release by checking agent components, controlling agent actions, and hunting through your AI security logs. Signals Corps Control Plane opens to early sign ups in the next few weeks, so join the waitlist with your company email.