Teams and audit log
A team sets what its members' AI coding agents may do in the team's repositories, and keeps a log of what they tried. Student and competition teams use it for free.
Create a team
On the Teams page, name the team (1 to 60 characters) and create it. Anyone can create one.
| Free team | Team plan | |
|---|---|---|
| Seats, the owner included | 5 | 10 |
| Members get the Team plan's limits | no | yes |
| Rules, repositories, audit log | yes | yes |
A team has 10 seats while its owner is on the Team plan, and 5 otherwise. One person can own or belong to 2 teams at most. With two teams, switch between them at the top of the page.
Members and roles
Add members by GitHub username. Someone who hasn't signed in to dagsec yet appears as invited, and joins when they first sign in.
| Role | Can |
|---|---|
| Owner | everything below, plus make members admins, turn email alerts on or off, and delete the team |
| Admin | add and remove members, add areas and repositories, put members in areas, change the policy, read the audit log |
| Member | see the team and its policy, switch their own area when allowed, and leave |
Areas
A large team can split into areas, such as frontend, backend and data. Each member works in one area at a time; an area's rules add to the team's and can only make them stricter.
- The owner or an admin puts each member in an area.
- They can also tick Can switch for a member, who then moves between areas themselves under My area. Owners and admins can always switch.
- A repository can belong to an area. Work in that repository always follows that area's rules, whatever area the member picked.
A team has up to 20 areas. Deleting an area keeps its members and repositories in the team, without an area.
Larger teams
Need more than 10 seats, for a team of 30 for example? Write to hello@dagsec.net and we'll add seats to the team.
Repositories
A team's rules apply to the work in its repositories. Add a repository as owner/repo (or its GitHub URL), or a whole GitHub account as owner. Each repository or account belongs to one team only.
When a member's AI agent runs a command, dagsec picks the team like this:
- A repository the team added, or a repository under an account it added, picks that team.
- When the agent only reports the folder it works in, a folder with the same name as a repository the team added picks that team.
- Anywhere else, a team that added no repositories applies, as it applies everywhere. If all of the member's teams added repositories, the member's own settings apply.
A team without repositories therefore works as it always did: its rules cover all its members' work.
Rules for AI agents
Owners and admins write the team's own rules on the Teams page; members read them. Each area can have rules of its own, which add to the team's for members working in that area.
| Kind | What it checks | Example |
|---|---|---|
| Command | a regular expression over every command an agent runs | jetson_clocks |
| Package version | a package an agent installs must stay in a version range | pypi torch >=2.3,<2.4 |
| Code pattern | a regular expression over the code an agent writes, in the files you name | /sys/class/gpio in **/*.py |
| Protected files | files an agent may not write at all | **/*.dts, config/calibration/** |
Each rule either warns (the agent asks the developer first) or blocks. When rules disagree, block wins. An area's rules can only make the team's stricter.
Where they apply.
| Where | Rules checked | When it blocks |
|---|---|---|
| Every command a member's AI agent runs (Claude Code, Cursor, Gemini CLI, Codex, Windsurf, Copilot, the desktop terminal) | command, package version | the agent doesn't run it |
| Pull requests to the team's repositories (the dagsec GitHub App) | code pattern and protected files on the changed lines, package versions in the lockfiles | the dagsec check fails, with a "Team rules" table in the comment |
| Code Guard in the desktop app, after the agent writes a file | code pattern, protected files | the agent is asked to undo the change |
| The desktop app's pre-commit hook | code pattern and protected files on the staged changes | the commit stops |
Code Guard and the pre-commit hook check on your computer: the code never leaves it, only the rules come from dagsec. They apply the team's rules even if a member turned their own Code Guard off. Pull requests are checked for repositories the team added under Repositories.
Use in other teams. Copy saved rules into other teams you own or run; a rule with the same name there is replaced. Try the rules shows what a command or a file would trigger before you save.
Versions: >=, >, <=, <, ==, !=, ^2.3, ~=2.3.1 and 4.8.*, comma-separated. An install that doesn't name a version also fires the rule, so the agent pins one. File globs work like .gitignore: ** crosses folders, and a name without a folder matches anywhere.
Rules only read what an agent runs or writes. dagsec never connects to your boards or machines.
Notes for the AI
The team, and each area, can write notes for Claude Code: the versions and services the code runs on, and how the team works. When a member starts a Claude Code session in one of the team's repositories, Claude starts with the notes and the list of rules in force, so it follows them from the first command instead of being stopped later.
- Up to 4,000 characters per team or area. The owner and admins write them; members read them.
- Notes that contain something that looks like a password, token or key are refused.
- Claude Code gets them through the desktop app's hook (connect Claude Code on the Integrations page). Other agents still follow the rules, but don't get the notes yet.
Suggesting rules
Members suggest a rule with Suggest a rule under the team's rules, for the whole team or an area, with a line on why. The owner and admins get a notification and approve or reject it; an approved rule joins the set right away, and the member is told either way.
Weekly email
Owners and admins get a line per team in their weekly email: how many AI agent actions the team's rules stopped, and how many they asked about.
AI agent install policy
The owner and admins set one policy for every member's Claude Code plugin, Cursor hook, Gemini CLI hook and MCP checks:
| Setting | Options | Default |
|---|---|---|
| Stop installs of packages with | critical and high vulnerabilities, critical vulnerabilities, or nothing (off) | critical and high |
| Also stop packages that don't exist | on or off | on |
| Also stop installs dagsec can't check | on or off | off |
| Stop packages under these licenses | comma-separated SPDX IDs, up to 20, such as GPL-3.0, AGPL-3.0 | none |
| When an install is stopped | block it, or ask the developer | block |
Installs dagsec can't check. By default dagsec fails open: if a package registry or the vulnerability database doesn't answer, or the month's agent checks are used up, the install goes ahead so an outage never stops your team's work. Turn this setting on to stop those installs instead (or ask, in ask mode). It applies to everything dagsec's server decides. It can't cover a developer whose machine can't reach dagsec at all, such as an agent with no network or a Claude Code session whose MCP server isn't connected; the agent runs the command in that case.
The team's setting overrides each member's own plugin mode. License matching follows the rules in License policy. Members see the policy on the Teams page but can't change it.
Audit log
Every install a member's agent attempts under the team's rules is logged:
| Column | Example |
|---|---|
| When | 5m ago |
| Member | @alice |
| Area | Backend |
| Packages | npm lodash@4.17.4 |
| Decision | Allowed, Blocked or Asked |
The owner and admins see the log. Entries are kept for 90 days.
Delete a team
Only the owner can delete a team. Deleting it returns every member to their own settings and plan, and deletes the audit log.