Skip to main content

Audit

Audit shows what happened in your organization: who did what, and when. Everything here is read-only.

Open it from the sidebar: select Audit. It is in the sidebar when you are an org admin, when your role may view projects (for Merge lanes and Memory), or when your plan includes Audit or Session transcripts and your role may view audit activity or sessions; see Subscription for what your plan includes.

Activity is a plan feature (Audit), and so is Sessions (Session transcripts). When your organization's plan leaves Audit out, the Activity tab still shows, and its panel says Not on your plan, as does a link to one Activity action: org admins get a See plans link; everyone else sees Ask an admin to change the plan.

Not on your plan: your plan doesn't include audit, with See plans

The tabs​

You see only the tabs your role allows. Audit opens on the first of them: when your role may view sessions but not audit activity, that is the Sessions tab.

The Audit tabs for an org admin on a plan without Session transcripts: Activity, Automations, Merge lanes and Memory

TabWhat it shows
ActivityEvery action taken in your organization, and who took it. Shown when your role may view audit activity, as org admins can. Off the plan its panel reads Not on your plan.
SessionsEvery agent session and its transcript. Shown when your plan includes Session transcripts and your role may view sessions. See Sessions.
AutomationsEvery automation in your organization, the org-wide list. Shown to org admins. See Automations.
Merge lanesEvery merge lane row of every project you may view. Shown when the merge lane is part of your edition and you may view projects.
MemoryThe memory units of every project you may view whose plan includes project memory. Shown when you may view projects and your plan includes project memory; off the plan, only when you may also edit projects.

Each tab has its own address, so a bookmark or a shared link opens the same tab. Times are shown, and filtered, in your time zone; hover From or To to see which one.

Activity​

Each row is one action.

Type in the Search box (Search or filter activity) and press Enter to find actions whose summary, person, path or kind contains your text; the search shows as a chip. It needs at least 3 characters: shorter text shows Type 3 or more characters to search. Press Escape to clear what you typed. As you type, the box also suggests these filters, each added as a chip:

FilterWhat it does
UserShows only these people's actions. System shows actions Jaah took on its own.
ResultShows only requests that ended one way: a class such as Client error (4xx), or one code such as 403 Forbidden.
RecordShows only actions on one record: a task, project, node, credential, account, user, group, cron or tag.
SessionShows only the changes made in one browser sign-in. Added from an action's details (All writes in this session), or by typing the session's id.

Select × on a chip, or press Backspace in an empty box, to remove it; press Enter in an empty box to drop the search text. Type shows every action (All), only Views, or only Changes. The list opens on Changes, or on All in an organization with HIPAA turned on. From / To limits the list to a time range. On a phone, and beside an open action's details, both sit under the Filters button beside the search, as Type and Time range. Clear filters resets them all.

While Result is set, a note says Result filters match web requests only. Views, pipeline events and system records are hidden. They come back when you remove it.

The search and filters are part of the page's address, so a bookmark or a shared link opens the same view.

Each row shows a mark for how the action ended, what was done, the action's name, who did it (System for an action Jaah took on its own) and how long ago; hover the time for the exact moment. Where the column naming who did it is hidden (on a phone, or beside an open action's details), a small mark before the sentence shows it instead: the person's initial, or ⚙ for Jaah itself. A refused sign-in for an address with no account names no one: it shows Unattributed, dimmed. An action taken with an API key shows Key and the key's name before the key's owner.

Rows the task pipeline wrote one after another are folded into one row; select it to see them all. One person's identical profile changes in a row fold the same way, their sentence counted, such as ×3. A sign-in and its sign-in record show as one row. Below the list, Load older adds earlier events, and a line says how many there are. For a long list the number is approximate, such as About 2.3M events or 10,000+ events. If earlier events fail to load, a banner gives the reason and Showing the last result that loaded.; select Retry to try them again.

An action's details​

Select a row, or Details in its ⋯ menu, to open its details in a pane beside the list; on a narrow screen the pane fills the screen. The pane's header names the event (Audit › Event and its number) and what was done, with the record it touched under it when the sentence does not already name it. The event has its own address, so you can copy the link and share it.

TabWhat it shows
EventWhen it happened, the Actor (with a copy mark for their id; for an action taken with an API key, the owner via API key and the key's name), the Action (with its short name under it, and a copy mark), its Type (read or write) and, for a request, its Result (the status it ended with). For a change, also the browser Session it came from (with All writes in this session), the Agent session an agent's browser named, the Browser, the Address our load balancer saw, and any Client-declared hops the client named before it, which Jaah cannot check. Select the copy mark beside an id to copy it.
RequestThe Method, the Path and the Status it ended with.
SubjectThe record the action touched: its Type, Id and Name. A record that was deleted, or that you cannot see, shows not resolvable (deleted, or not visible to you).
PayloadEach field the action recorded; a long value is cut short and marked (truncated — see the full payload below). Below the fields, request body and response body (when recorded) and Full payload (JSON) open the full JSON.

A tab appears only when the action recorded something for it.

Select the × (Close pane) at the pane's top right to go back to the list; the arrows beside it (Maximize pane) widen the pane. A link to an event that does not exist says This event link is not valid.

Empty states​

MessageMeaning
No activity recorded yetNo action has been recorded.
No activity matches your filtersNo row matches your filters. Select Clear filters to see them all.
Could not load the activity logThe list couldn't be read; the reason follows. Select Retry.

Merge lanes​

Each row is one pull request in a merge lane, newest change first. A mark at the start shows its state.

ColumnWhat it shows
PRThe pull request's number, the title of the task that opened it, and why it last moved or why it failed, such as Tests failed. Select the row to open that task.
Forge↗ opens the pull request on its forge.
ProjectThe project's name. Select it to show only that project.
AgeWhen the row last changed. Hover it for the exact time.

Load more at the bottom adds older rows. Before any lane has handled a pull request, the list reads No merge lane rows yet.

Audit, Merge lanes: the Project and State filters above merged pull requests, each with its task title, Forge, Project and Age

FilterWhat it does
ProjectShows one project's whole merge lane (below). Selecting a row's project name does the same.
StateAll states, Open (queued, testing or failed) or Settled (merged, superseded or closed). Shown only with All projects. With no row in that state, the list reads No open rows or No settled rows, with Clear filters.

With a project picked, the tab shows that project's merge lane. It reads the lane's work and updates on its own while the lane runs. Its tabs, each with a count, are Queue, Runs, History, Certificates, Flakes and Quarantined. Each row is a pull request, a run, a certificate or a test, with its state; select a row's task to open it. Runs lists each run's Batch (what it merged, or the commit it checked), Merge (start to last merge), Gates, Wall (wall time) and Age, 20 to a page with Previous and Next. An empty tab says so, such as Nothing queued. or No runs yet.

One project picked: Queue, Runs, History, Certificates, Flakes and Quarantined with their counts, and the lane's dot at the end

The lane's state is a row of icons at the end of its tabs: point at one to read it, or select it to keep it open with its links.

IconMeans
DotActive (green), Held or Draining (amber), or Off (hollow). While a run is live, selecting it opens that run's console. Off says why, with a link to the setting that turns it on.
WarningThe lane is slow: a pull request has waited, or nothing has merged, longer than the limit, or a base check is outdated (linked to its task). See Runs opens its runs.
CrossThe integration branch is red: the tests failing on it, the gate log and the task fixing it.
InfoThe branch was red at a commit, and has not been re-checked since its fix was done or it moved on.
ServerA node the lane's runs are kept off until a time, why, and what to fix before then.
MergePull requests set to merge on their own outside the lane: they merge untested.
StopStop lane, while the lane drains (below).

Held means a lane run is stopped (or asking, or reset) and no other run starts while it stands. The dot names the run: open it, then Start it to resume the lane or Complete it to let the next run start. Stopping a single run from its task's ⋯ menu holds the lane this way.

Draining means the lane was switched off but still holds a run or queued pull requests: they merge first. For someone who may edit the project, the Stop lane icon ends that now. It stops the run, empties Queue and tells each runner to merge its own pull request; an icon then says how many runs it ended, pull requests it closed and runners it told, until you select Dismiss on it. Anyone who may edit the project's tasks can also stop a single run from its task's ⋯ menu.

What the lane does, which plans include it and how to turn it on: Use the merge lane.

Memory​

What runners have learned, kept as units that later tasks can draw on. Each row is one unit, newest first, with the name of its project; select the name to show only that project. Load more at the bottom adds older units. The Project filter picks one project. The Project memory setting in the project's Settings › Memory decides whether its tasks use the units.

Each row is one unit. A mark at the start shows its status; hover it for why the unit has that status, such as contradicts canon directive … or stale at …. Then its ID (such as mem-12), its title, and:

ColumnMeaning
ProjectThe unit's project. Select it to show only that project. Only on the all-projects list.
KindThe kind of unit, such as gotcha or codemap.
TrustHow far the unit is trusted, from 0 to 1.
InjectedHow many tasks had the unit in their prompt.
SeenHow many tasks were shown it while they worked, or when their session started.
UsedHow many tasks read it or cited it in a merged PR or summary. A task saying it used the unit doesn't count.
WrongHow many tasks flagged it wrong.

Hover a count for what it counts. Select the arrow at the start of a row to open it: every fact above, its Repository on a project with more than one, its Status and Reason, its Source (the task that taught it, with a link to its transcript, or what produced it) and the unit's whole text, including its Repro: line.

The row's ⋯ menu, shown only to someone who may edit that unit's project:

ItemWhat it does
ReleaseOn a held unit: lets tasks be given it, without waiting for the canon check.
RetireStops the unit being used. Not on a retired unit.
PurgeDeletes the unit for good and removes it from every stored runner prompt, after you confirm. Use it for sensitive data; otherwise retire. Each purge is recorded in Activity (who, which unit, how many prompts were scrubbed).

Before any runner has learned something, the list reads No memory yet. Off the plan, the all-projects view shows one line: Memory (what runners learn on your projects) isn't in your plan. and where to turn it on.

Memory with All projects, off the plan: the line linking Billing

With a project picked, the tab shows that project's whole memory view. The chips at the top say whether each part of memory is working: Completion, Stop delta, Code map, Distill, Secret check and Canon check. The running ones share one chip, such as 4 active; select it to see them. Any that isn't running gets its own chip reading Inactive, Rules only or Off; hover it to see why. When the memory model is missing, Set memory model beside the chips opens the setting. A running Code map chip shows, in green, which version of the code map is built, or not built in grey before the first build. Held by canon check counts, by feed, the accepted units the canon check keeps from runners until a memory model is set; each reads held in the list. It shows only when some are held; select the count to list them.

ItemMeaning
SearchFinds units by title, text or number, such as mem-12.
KindShows one kind of unit: codemap, gotcha, recipe, env or qa.
StatusOpens on Live: every unit still in play. All shows every unit; or pick one status: Proposed, Accepted, Flagged, Stale, Retired or Graduated. Held shows the units the canon check keeps from tasks. In the list, an accepted unit that tasks can be given reads injectable, and one the canon check keeps from them reads held; one given to 20 tasks that none used reads unused and is no longer given to tasks, though runners can still search it. A gotcha proposed without a repro reads needs repro and can only be searched.
SortOrders the list: Newest first, or the most Injected, Seen, Used or Wrong first.
RepositoryShows the units of one connected repository. Only on a project with more than one repository.

Under Newest, the chunks of one code map fold under their first, marked with how many more there are, such as +3; open the row to list them. Sorted by a count that no unit has yet, a line says so, such as No unit injected yet. With nothing to list, the view reads No matching units when a search or filter is set, No live units on Live (with Show all), or No memory yet.

When the organization's plan doesn't include project memory, a picked project's view says so in one line and names where to turn it on: Billing for an org admin, otherwise An admin can turn it on. For someone who may edit the project it then reads Delete a memory unit for good. Task reports show its ID, such as mem-12. and holds one field, Unit ID: type the unit's ID and select Purge to delete it for good after you confirm, so stored memory can always be erased.

The Memory view with a project picked, without project memory in the plan: the Billing line, and Unit ID with Purge