Feature Flag Governance: What an Auditor Will Ask For

Feature flag governance for teams with an auditor: change reasons on production, access in the audit trail, required 2FA, token caps and environment diffs.

  • governance
  • audit-log
  • security
  • compliance

The auditor’s sample had eleven production changes on it. Ten were deploys. Each came with a pull request, two approvals and a pipeline run stamped to the second. The eleventh was a flag that switched on a new payment path in March. It went fine, and everyone remembered that. Nobody could show who turned it on, why that day, or whether the person still had access, because the deploy pipeline had been answering those questions for years while the flag dashboard grew up beside it and was never asked anything.

Feature flag governance puts flag changes under the same rules as the rest of production. That covers who may change what, a written reason for each change, a record of who had access, protected accounts, and a way to show production matches what you tested. Below are the five questions that come up in a SOC 2 or ISO 27001 review, with the Featureflip control and the record that answers each.

Key Takeaways

  • Auditors sample flag changes the way they sample deploys. A flag flip needs the same evidence: who, what, when and why.
  • Make production demand a reason, and enforce it on the server so API scripts can’t skip it.
  • Keep membership changes in the same log as flag changes. An access review then reads one record.
  • Require 2FA for everyone who can change production, and put an expiry on every personal API token.
  • Compare staging with production before a release, and copy targeting across after a preview.
  • Match audit log retention to your audit period, or stream flag changes somewhere you keep longer.

1. Why flag changes slip past change control

A deploy leaves a trail without anyone trying. The pull request names an author and a reviewer, CI records what ran, and the pipeline stamps the moment the build reached production. That trail is the evidence SOC 2’s change management criterion (CC8.1) looks for when it expects production changes to be authorized, tested, approved and documented.

A flag flip skips all of it. One click can turn on code that has sat dark in production for three weeks, and customers feel that change exactly as they would feel a deploy. Once a reviewer learns that your releases go out behind flags, the flag platform joins the sample, and “the code was reviewed when it merged” stops working as an answer for something that happened weeks after the merge.

2. Five questions an auditor will ask about your flags

Auditors word them differently, and ISO 27001 files them under other control numbers than SOC 2 does, but these five come back every time.

Five audit questions mapped to flag controls A three-column table. Who can change production is answered by roles and critical environments. Who changed it and why is answered by required change reasons. Who had access is answered by membership entries in the audit log. How accounts are protected is answered by required 2FA and token lifetime caps. Whether production matches what was tested is answered by compare and copy between environments. The auditor asks The control The evidence Who can change production? Roles, critical environments Role list, Critical badge Who changed this flag, and why? Required change reasons Reason on the audit entry Who had access, and who granted it? Membership in the audit log Added, removed, role entries How are those accounts protected? Required 2FA, token caps 2FA column, token list Is production what you tested? Compare and copy Diff, copy audit entry
Each question an auditor asks about flags, the Featureflip control that answers it, and the record you hand over.

The first question is partly about roles. Featureflip has four: a Viewer reads, a Member edits flags and targeting rules, an Admin also manages environments and members, and an Owner holds billing and organization settings. Keep the Admin and Owner lists down to the people who would answer for an outage. The team management guide has the full role table.

3. Make production ask why

Mark production as a critical environment. The setting has two switches.

Require a change reason blocks any change to what the environment serves until someone writes down why. The server enforces it. So the rule holds in the dashboard, on the Management API, through the MCP server, and in any script your team wrote against the API. A request without a reason gets 422 with change_reason_required, and nothing is written. This switch is on the Business and Enterprise plans.

Require typing the environment key asks for the key, usually production, before the dashboard saves. It catches the engineer who thinks they’re in staging. Every plan has it.

Edit a segment that rules depend on and every environment using it serves the new version at once, including production, even if you opened the segment from a staging page. So Featureflip asks for a reason on segment edits, variation edits, and flag archives or deletes as soon as any environment in the project requires one. Flag names, descriptions, tags and owners stay exempt.

Any member who can edit the environment can turn a switch on, while turning one off takes an Admin or Owner plus a reason, so the audit log shows who relaxed production and what they said. New projects start with production already guarded.

4. Write reasons someone can audit a year later

A required field invites the laziest answer that passes, and “fix” passes. An auditor reading it ten months later learns nothing from it, and neither does the engineer on call next spring.

We’d lead with the ticket or incident ID, then the cause. “INC-412: rolling back after the 14:02 error spike” takes six seconds to type and survives any audit, since it points at a record with its own trail. A team convention plus a glance at the audit log during your access review keeps the quality up.

Automation is where turning this on goes wrong. A deploy script or CI job that changes production flags through the API has to send the reason too, in the X-Intent-Reason header:

Terminal window
curl -X POST "https://api.featureflip.io/api/v1/orgs/acme/projects/checkout/flags/new-checkout/environments/production/toggle" \
-H "Authorization: Bearer $FEATUREFLIP_TOKEN" \
-H "Content-Type: application/json" \
-H "X-Intent-Reason: INC-412: rolling back after the 14:02 error spike" \
-d '{"enabled": false}'

Grep your pipelines for calls to the flag API before you flip the switch, or the next scheduled job fails on a 422. The flag cleanup Action sends the header already, from v2.6.0, and names the pull request that removed each flag it archives.

Scheduled changes keep the reason they were created with. A rollback queued on Monday for Thursday night lands with Monday’s words attached.

5. Keep the access record next to the change record

In a quarterly access review, someone lists everyone who could reach production systems, checks each person against their current job, and writes down what was removed for SOC 2’s logical access criteria (CC6.2 and CC6.3). For flags, that used to mean a screenshot of the members page. And memory.

Featureflip’s audit log now keeps that history beside the flag history. Members added by accepted invite, new organization or single sign-on provisioning are there, along with removals, every role change with the old and new role, and invites as they’re created, resent, accepted or revoked (the link itself is never stored). So is 2FA. A member turning it on or off shows up in every organization they belong to, and service tokens are logged with the role they were granted. Provisioned members carry SSO provisioning as the actor.

Filter by entity type and date range. Removing an SSO member in Featureflip also revokes their API tokens, which removing them at your identity provider alone doesn’t do.

Then there’s retention. Featureflip keeps entries for 7 days on Free, 30 on Team and Pro, 90 on Business and 365 on Enterprise, while a SOC 2 Type II observation period can run as long as twelve months, with the sample landing anywhere inside it. If your plan’s window is shorter than your audit period, pick a longer one, or point a webhook at your own store so every flag, rule and SDK key change is kept as it happens.

6. Lock down the accounts that can change production

Every reason in the log is signed by an account, so the account has to belong to the person it names. Two settings on the Security policies page in Organization Settings deal with that, one for how people sign in and one for their API tokens.

Requiring two-factor authentication comes with Team and above, with a grace period of 0, 7 or 14 days during which members without it see a banner counting down to their deadline. After that they get a setup screen. Nothing else loads. Their personal access tokens stop working too, with two_factor_required on the API, which closes the path through Terraform and the MCP server that a dashboard-only rule would leave open. Nobody can switch 2FA off while the organization requires it, and the Owner flipping the switch needs it first. Members who sign in through your enforced single sign-on connection are exempt because their MFA lives at your identity provider. Service tokens are exempt too. An organization has no phone to read a code from.

A downgrade to Free leaves the requirement in force until an Owner turns it off.

The maximum token lifetime is on Business and above. An Owner caps new personal access tokens at 30, 90 or 365 days, and tokens that never expire can’t be created at all while the cap applies. Older tokens over the limit keep working at first, listed on the same page with their owners. Ask those people to replace them before you turn on Refuse existing tokens over the limit, or a forgotten cron job finds out at 2 a.m. After that, the leftovers get 403 token_lifetime_exceeds_policy from your organization. Nothing is revoked.

7. Show that production matches what you tested

“Tested in staging” is a claim about two environments being the same. An auditor following CC8.1 will want to see it.

The Compare page lists every flag whose on/off state, default variation, strategy, rules or prerequisites differ between two environments. It’s on every plan. Rule order counts as a difference, since it decides which rule matches first, and a flag with no configuration on one side also shows as different, so nothing forgotten slips through.

Copying is on Team and above. Pick a flag and a source environment, and a preview shows what will change in the target before anything is saved. It carries the default variation, strategy, targeting rules and prerequisites. On/off stays put unless you tick the box. Prerequisites get checked in the target first, and a missing or archived one refuses the whole copy with each reason listed. If either environment changes after the preview, Apply is refused and the preview runs again. Buckets survive too. The bucketing salt belongs to the flag, so a copied percentage rollout puts the same users in the same buckets.

For the audit, the result is a summary entry naming the source environment, next to the individual changes and the reason a critical target asked for. “Copied from staging after review, CHG-2231” reads cleanly to an auditor. The compare and copy guide covers the API endpoints and MCP tools.

8. Which controls come with which plan

ControlFreeTeamProBusinessEnterprise
Type the environment key to confirmYesYesYesYesYes
Compare environmentsYesYesYesYesYes
Copy a flag between environmentsYesYesYesYes
Require 2FA for every memberYesYesYesYes
Required change reasonsYesYes
Personal access token lifetime capYesYes
Single sign-onOIDCOIDC, SAML, SCIM
Audit log retention7 days30 days30 days90 days365 days

A small team on Team can cover most of the five questions with 2FA, the typed key on production and a compare before each release. The written reason and the token cap arrive with Business, along with single sign-on through your own identity provider. Plans and limits has every number in one table.

For flags with a large blast radius, such as auth, billing or anything touching most users, have a second engineer pair on the change and put their name in the reason. The audit entry then carries both people. Our feature flag best practices guide covers that habit alongside the role split it depends on, and the audit trail overview shows how the log fits with the rest of the platform.

The shorter version

Frequently asked questions

What is feature flag governance?

It’s the set of controls that treat a flag change as a production change. In practice that means limiting who can change production flags, recording who changed each flag and why, keeping a history of who had access, protecting those accounts with 2FA and expiring tokens, and checking that production’s targeting matches what was tested.

Are feature flag changes in scope for a SOC 2 audit?

Plan as if they are. SOC 2’s change management criterion, CC8.1, covers changes to production systems, and a flag flip that changes what production serves is one of those, so expect flag changes in the sample next to your deploys.

What should a feature flag audit trail record?

Four things per change: the actor, a timestamp, the before and after values, and the reason. An access review also wants membership changes, role changes and invites in that same log, so one filter answers who could do what during the quarter. Featureflip records all of it. Each entry keeps the request’s IP address and user agent too.

How do you require a reason for every production flag change?

In Featureflip, mark the environment as critical and switch on Require a change reason (Business and Enterprise plans). The server then refuses any change without one. Dashboard users get a prompt. API calls and scripts send the reason in the X-Intent-Reason header, and a missing header returns 422 change_reason_required with nothing written.

How long should you keep feature flag audit logs?

At least as long as your audit period. A 30-day log can lose most of a SOC 2 Type II observation period before the auditor picks a sample. Featureflip keeps 90 days on Business and 365 on Enterprise, and webhooks can copy each change to storage you control.


Featureflip is a flat-priced feature flag platform with critical environments, an audit log that records access alongside flag changes, required 2FA, and compare and copy between environments. Start on the Free plan without a credit card. Change reasons and token lifetime caps come with the Business plan.