CodeCargo logo

Platform

Audit Log

The Audit Log is a reverse-chronological record of changes made in your organization — who acted, what they acted on, when, and whether the action succeeded, was denied, or errored.

Find it under Govern → Audit Log in the organization sidebar.

Plan availability

Viewing and exporting the Audit Log is an Enterprise plan feature. Events are recorded for your organization regardless of plan, so when the log becomes available it opens onto the history your organization has already accumulated rather than an empty table. Talk to us to turn it on.


Reading the Log

Each row shows the time of the event, the action, the actor, the target, and the outcome. Actions, module names, and target types are shown exactly as they are recorded, so the value you see is the value you can filter on or query in a SIEM after exporting.

Denied and Error outcomes are badged so you can scan for them without reading every line. Actor names are captured at the time of the event, so a renamed or removed user still shows who acted.

The Audit Log listing CargoWall and Platform events with time, action, actor, target, and outcome; a denied call and an error are badged

Expand a row to see the field-level diff for the change — old value on the left, new value on the right — alongside the detail that has no column of its own: module, event ID, target ID, source IP, trace ID, and user agent. A field that was never set is shown as —, and one explicitly set to an empty string as (empty).

An expanded cargowall.policy.updated row: the changes to name, description, and defaultAction as old and new values, then the module, event ID, source IP, trace ID, target ID, and user agent

What's Recorded

The log records successful changes across your organization: access and roles, projects and repositories, building blocks, guardrails, CargoWall, agents, Dispatch, migration, notifications, operators, and organization settings. A change is recorded the same way whether you make it in the app or an agent or MCP client makes it as you.

Each row names who acted and what they acted on. When the previous value matters, the row also includes a before-and-after of the fields that changed:

  • Guardrail rules — saves, deletes, and turning a rule on or off — and switching guardrails on a repository, workflow, or project
  • Access — organization and team roles, project access, self-service workflow access, and building block shares
  • Organization settings — support access, LLM spend controls, and runner rates
  • Agents and MCP servers — tools, approvals, and publishing or retiring
  • CargoWall policies and baselines
  • Credentials — which secret or token was set, replaced, or cleared, never the value
  • Migration Assistant source connections — settings, linked organizations, groups, or servers, granted projects, and repository selection, plus the organization's dry-run switch
  • Migration plans — starting and canceling the plan, the approach, features, copy rules, rules and their order, decisions, close-out items, a project's destination or wave, a repository's name, and a plan wave's name and dates; and a wave's, batch's, or pattern's name, description, and settings. A setting stored as a group names only the part that changed, such as features.conversion.

A row without that before-and-after still expands to the module, event ID, and the other details above. A change that leaves things as they were is not recorded, and a failed call is not recorded as a success.

Some writes are left out so they don't bury the actions the log is for:

  • Your interface preferences, edits to files in a workspace, and an unsaved guardrail draft you discard
  • Chat messages, and stopping your own in-flight reply. Creating, renaming, or deleting a thread is recorded
  • Traffic from the importer while an import runs. Starting, finishing, failing, stopping, or abandoning the import is recorded
  • Deleting the organization, which removes the log with it

Action names are stable dotted labels, such as projects.project.created. A filter or a SIEM query can rely on the name shown in the row.

Migration Assistant Background Work

Some Migration Assistant work finishes after the request that started it. Its rows name the person who started or queued that work as the actor, even when a later check records the row. When no one in the organization did, the actor is the system.

ActionRecorded when
migration.repository_migration.finishedA repository migration ends, with how many repositories came out each way. Each repository is a field: its GitHub copy, the step it reached, and its failure category
migration.repository_migration.source_window_opened, .source_window_closed, .source_window_left_openA repository disabled in the source is enabled for its copy, disabled again with its permissions put back, or could not be put back
migration.repo_relocations.migration_recordedA migration's copies record new or changed relocations. The row counts any that replace a relocation a person confirmed
migration.update_pull_request.opened, .merged, .closedCodeCargo opens, merges, or gives up on and closes an update pull request. A pull request you merge on GitHub yourself is not recorded
migration.rewire_decision.reopenedA question the scan asks, the AI review, the AI sweep, or an import replaces a person's decision on a Prepare for Rewire finding. Filed on the organization
migration.import_run.abandonedAn import is abandoned. When an idle import is abandoned on its own, the actor is the system; when someone abandons one, for example from an MCP client, the actor is that person
migration.import_run.repositories_activatedAn import activates repositories it linked
migration.source_job.endedA job against a connected source ends. The job's message is not recorded

Source Connections and the Dry-Run Switch

A connect, a check, a disconnect, or a settings save for Azure DevOps, GitLab, GitHub Enterprise Server, or Bitbucket Data Center is filed under that connection. Filter the action, such as migration.gitlab.group_disconnected or migration.azure_devops.settings_updated. The target type is SourceConnection and the label is the host, so a search for the host finds the rows.

The row shows each changed setting, old value to new: status, the repository naming pattern, the archive storage host, whether an organization, group, or server is linked, and Azure DevOps projects granted or given up. When they change, an Azure DevOps connection names the tenant and service principal, and a GitLab connection names the account an OAuth grant acts as.

A credential is only set, replaced, or cleared, by kind. The token, password, or key is never recorded. A Bitbucket password connect names the account, not the password. A GitHub Enterprise Server connect names the organizations it will link; an app connect also names the app id, never the private key. When a check or connect starts a job, the row names that job. How the job ended is migration.source_job.ended. Starting a sign-in with Microsoft, GitLab, or Bitbucket writes nothing on the connection, so that call is recorded without a before-and-after.

Choosing which repositories a connected organization or group keeps (migration.source.repository_selection_set) is filed under the same connection. Each repository added or removed is named by the path you see when you choose it, or by id when nothing lists it yet, and the row counts how many were added and removed. Switching from every repository to a list names the ones the list leaves out. Switching to every repository records only that. Saving the same selection records nothing.

Turning the organization's dry-run switch on or off (migration.settings.updated) is filed on the organization, with dryRun from the old value to the new one.

Filtering

Each column header has a filter menu: Action filters by module (for example, all CargoWall events) or by individual action, Actor by who acted, Target by target type, and Outcome by Success, Denied, or Error. The search box does a substring match on actor and target labels.

The filter menus only offer values that actually appear in the selected time range, so a filter never returns an empty result.

The time range selector sits in the page header and defaults to the last 30 days. Widen it to reach older events — the log retains more than the default window shows.

Every filter and the time range live in the URL, so a filtered view is a link you can share or bookmark. The first page loads up to 100 events; use Load more to append the next page.


Exporting

Export CSV downloads the full filtered match set — not just the rows currently on screen — as audit-log-YYYY-MM-DD.csv. The file includes every column the log records, timestamps in ISO-8601 UTC, and the field-level diff, so it can be correlated with other logs in a spreadsheet or SIEM.

Very large exports are capped. When that happens the file still downloads and a warning tells you how many events it contains; narrow the time range or filters to export the rest.

Export is separately permissioned. If you can read the log but not export it, the button isn't shown — ask an organization administrator if you need it.