How Tego works
Five steps, one loop. Scroll, and each one builds itself.
1
Write your rules once.
One short rules file (agent.yaml) says what the agent may read, change and call. Writes outside the folders it lists are refused. Hosts and programs are limited once it lists them.
- Readyour codeAllowed
- Changefiles in
./srcAllowed - Call
api.anthropic.comAllowed - Ask first
./confidentialAsk me - NeverSSH keys,
.envNever
2
Run your agent as it is.
Claude Code, Codex or your own agent starts under Tego. No code changes. A ready rules file sets the few settings each agent needs.
$ tego run agent.yaml
- Read
src/billing.pyAllowed - Edit
src/billing.pyAllowed - Run
pytestAllowed
3
Wrong moves stop. The work goes on.
Tego refuses what your rules rule out and tells the agent why, so it can try another way.
- Upload to
pastebin.comBlocked - Write
~/.bashrcBlocked
claude was told why, and kept working.
4
Ask first? Tego asks you.
Your agent’s own prompts can be switched off. This one can’t: Tego asks from outside the agent, whichever agent you run. Allow it once or for the whole run, or deny it. Either answer can become a rule for the next run. For a repository you don’t trust, still use a VM.
claude started with --dangerously-skip-permissions
./confidential/plan.mdAllow onceDenyMore choicesYou allowed it once. Your answer is in the record.
5
Every run is saved.
Programs, writes and connections, with the rule behind each. Check it later: an edit by the agent shows.
- Upload to
pastebin.comrule: call only Anthropic’s APIBlocked - Read
plan.mdyou said yesYou allowed - Run
pytestrule: read your codeAllowed
Record verified. The agent changed nothing.
Try it on your machine today.
curl -fsSL https://github.com/Attention-Span-AI/Tego/releases/latest/download/install.sh | sh