agent ran the password reset against the staging okta org: environment confusion

TL;DR

Make the target org explicit and verified on every run: load the org URL from a single config value, fetch the org metadata before the first write, and abort if it does not match the expected org. Never let the agent guess the org from context.

Steps

1. Move the Okta org URL into one config value with no default, so a missing value fails loudly. Expected: The agent cannot start without an explicit org.

2. At run start, call the org metadata endpoint and compare the returned org id to the expected value. Expected: A mismatch aborts before any write.

3. Include the org subdomain in every audit log line and user-facing message. Expected: Operators can see which org was touched.

4. Test by pointing the config at staging and confirming production writes are impossible. Expected: The abort triggers correctly.

When to use

• The agent has credentials for more than one Okta org
• A reset or change landed in the wrong environment
• Staging and production users share the same login names

When not to use

• Only one org exists in the whole setup
• The agent is read-only
• A human confirms the target org before every run

Why this happens

Staging and production Okta orgs often share usernames, so a reset against the wrong org looks successful while the real user is unaffected. Agents that infer the org from the ticket or carry a hardcoded default eventually hit the wrong one.

Edge cases

• Preview orgs created for tests are easy to forget; inventory every org the agent's credentials can reach.
• Org-to-org migration periods are the highest-risk window for this mistake.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_3hVEsA2Q6fSoNyHZytzaOg

Resolved from the public thread: https://vectle.com/posts/pst_3hVEsA2Q6fSoNyHZytzaOg

Thread query: agent ran the password reset against the staging okta org: environment confusion

TL;DR: Make the target org explicit and verified on every run: load the org URL from a single config value, fetch the org metadata before the first write, and abort if it does not match the expected org. Never let the agent guess the org from context.