Security and privacy

This page documents security-relevant behavior and privacy controls in auto-agents.

Secret handling

auto-agents separates non-secret configuration from secret material:

Non-secret settings are stored in auto-agents.toml.
Secrets such as tokens, PEM keys, and local inference API keys are stored via OS credential vault/keychain mechanisms.

Do not commit secrets to repository configuration files.

Code review bot identity

code-review operations require a bot identity separate from the PR author to satisfy GitHub review constraints.

Recommended setup is a dedicated GitHub App with minimum required repository permissions.

Telemetry model

auto-agents integrates anonymous usage telemetry for product and reliability improvement.

Design constraints:

No collection of user prompt/content payloads.
No PII by design.
Endpoint and app identity are fixed in code.

Telemetry can be disabled via environment variables or auto-agents.toml.

Telemetry opt-out methods

Global opt-out:

export DO_NOT_TRACK=1

Application opt-out:

export AUTO_AGENTS_NO_TELEMETRY=1

Config opt-out:

[telemetry]
enabled = false

Local security review support

auto-agents includes a security-review action and security findings UI panel. Use these for routine checks during PR review and release preparation.

Practical security guidance

Prefer least-privilege tokens and app scopes.
Use short-lived credentials whenever platform support exists.
Rotate bot credentials on schedule or after exposure events.
Audit .github/labels.yml creation in repositories where policy controls metadata writes.

16 July 2026