Trent

Trent agent plugin

Trent, an AI security engineer. Review code, plans and configs for security problems, run threat models over a repo or website, and track remediation without leaving the editor.

Installing

Add the marketplace by repository name. trnt-ai/trent-agent-plugin is the only public marketplace that carries Trent, and adding it by name is what ties the install to us. (A plugin manifest carries no signature, so the name is what you are trusting; never add a marketplace from a URL to a file.)

Claude Code:

/plugin marketplace add trnt-ai/trent-agent-plugin

/plugin install trent@trent

Codex CLI:

codex plugin marketplace add trnt-ai/trent-agent-plugin

codex plugin add trent@trent

Check what you installed with claude plugin details trent. It prints Source: trent@trent and the manifest's identity fields.

Signing in

The plugin declares a remote MCP server and no credentials. Authentication happens in the browser the first time a tool runs; there is no API key to paste and none is shipped here.

What the plugin sends, and where

The plugin holds no code of its own: four skills and the addresses of Trent's two servers. Everything it sends goes through your coding agent, as Trent tool calls, to the server for your account's region: https://mcp.trent.ai/mcp (US) or https://mcp.eu.trent.ai/mcp (EU).

• Repository scans. Trent reads a repository through the Trent GitHub App you install on it. The plugin does not upload your repository; it sends the repository's owner and name.
• Reviews. security_advisor sends the content your agent passes it for that review: code, a diff, a plan or a config. /trent:trent-loop sends the diff of each fix it makes.
• Documents. A design document or findings file you choose to attach is uploaded straight to Trent's storage on Amazon S3, through a signed URL that prepare_file_upload returns and that expires.
• What you type. Feedback, edits to findings and project settings go as tool arguments.

The plugin also sends its version in the x-trent-plugin-version header. It reads no credentials from your machine.

Trent handles this data under its privacy policy, https://trent.ai/privacy-policy/. Security documentation is in the Trent Trust Center, https://trust.trent.ai/.

Using it

Ask your coding agent in plain words; the plugin's skills tell it which Trent tool to use:

• Review one piece of work (a diff, a plan, a config): trent-security-advisor.
• Scan a repository or website and read what Trent found: trent-threats, or trent-repo for the repository you are in.
• Fix what Trent found: run /trent:trent-loop, or /trent:trent-loop 3 to fix at most three. It works on a new branch, checks each fix with your own tests and with Trent's security advisor, and opens one pull request for you to review. After you merge it, run it again: Trent scans the merged code, reports which fixes it now counts as done, and carries on with the rest.

Verifying where a release came from

Each release is one commit, and its message carries the revision it was built from:

Published from source revision <sha>

Run git log -1 in a clone of this repository to read it. The is an identifier, not a link; there is nowhere to follow it to. What it gives you is an exact name for the build you are running: quote it to support and they can say which release you have. If a copy of this plugin does not carry a message in that form, it was not published by us.

Each release's pull request also runs a check, provenance, that the commit has that shape: one commit on main, that subject, plain files, and a changed version when the content changed. It reads the commit and holds no credential.

The plugin manifest names https://github.com/trnt-ai/trent-agent-plugin, this repository, as its repository, and trent.ai as its homepage. A plugin claiming to be Trent from anywhere else is not ours.

Pinning a release

Every release is tagged plugin-v , on the commit that release published. git tag -l in a clone lists them. A tag here never moves: the repository's rules refuse it, so a pin keeps resolving to the same files.

On a Team or Enterprise plan an admin can register this marketplace for everyone in managed-settings.json, pinned to a release. ref is the tag and sha is the commit it names; with both set the sha is the pin and the ref says which release it came from:

{

"extraKnownMarketplaces": { "trent": { "source": { "source": "github", "repo": "trnt-ai/trent-agent-plugin", "ref": "plugin-v0.1.0", "sha": " " } } }, "enabledPlugins": { "trent@trent": true } }

Read the sha off the tag rather than copying one from anywhere else:

git ls-remote https://github.com/trnt-ai/trent-agent-plugin refs/tags/plugin-v0.1.0 'refs/tags/plugin-v0.1.0^{}'

One line back means the tag names that commit: pin it. Two lines back means the tag is annotated: the line ending ^{} names the commit, and that is the one to pin.

Upgrading is then a deliberate edit of those two fields, not something that happens when someone reinstalls.

Getting help

Questions, bugs and feature requests reach us at https://trent.ai or through your usual support channel; that is where they get answered.

This repository is published from Trent's build on each release; every file in it is generated, so it is not the place to file or fix anything.

Licence

Copyright Trent AI. See LICENSE for the terms, and https://trent.ai for anything the terms do not answer.