Read the Agent audit log
How do I read the Agent audit log and find one action?
Sydekick → Agent → Audit logs holds the actions your agent has taken. Four tiles summarise the actions currently on the page, and the table below lists them, one row per conversation, approval or run. Narrow it with search or the Client, Capability, Initiator, Status and Date controls, open any row for its detail, and use Export to download the rows you are looking at.
Key facts
| Where it lives | Sydekick → Agent → Audit logs. Page access is set per team member |
| Tiles | Actions in period, Credits in period, Failure rate, Capabilities used |
| Filters | Search, Client, Capability, Initiator, Status, Date, Playbook runs only, Failed only, Show step detail |
| Date options | This month, Last 24 hours, Last 7 days, Last 30 days, Last 90 days. The log reaches back at most 90 days |
| One row is | A conversation (all its turns together), an approval request, a run or another action |
| Default view | Hides the automatic wrapper rows for playbook runs; Playbook runs only shows just those |
| Export | A CSV of the rows currently loaded and filtered |
Steps
Open Sydekick → Agent → Audit logs. If the page says to enable your agent first, see setting up Sydekick.
Set Date first. It decides how far back the log reaches.
Narrow with search, Client, Capability, Initiator and Status, or tick Failed only or Playbook runs only. The line under the controls counts the actions in view.
Select a row's arrow to open it. Turn on Show step detail to expand steps inside the rows.
If the page says only the most recent are shown, select Load more, or narrow the filters, before you conclude anything from an absence.
Select Export to download what you see.
Start by reading the tiles for what they measure
All four tiles are totals of the rows currently loaded on the page, under the filters you set, so they change when you filter and when you select Load more. They are a summary of the view, not a separate statement from the system.
- Actions in period counts the rows in view: conversations, approval requests, runs and jobs. A conversation with several turns is one action.
- Credits in period adds up the credits recorded on those rows. Its sub-label says the range, such as This month. It is not the account's credit balance and can be lower than the figure on Agent → Home, so use Usage & cost for what the account has used and what is left. For some actions the credit figure is worked out from the agent's recorded usage rather than read from the ledger, so do not expect the two pages to match to the credit.
- Failure rate prints its arithmetic underneath, for example 1 of 9 failed, and turns amber and then red as it rises.
- Capabilities used counts the distinct skills the agent used. It does not count the integrations you have connected; those are on Capabilities.
This month here means the calendar month starting on the first, in UTC, not your billing period. That is another reason these tiles differ from Home on a plan that renews mid-month.
Narrowing it down
The search box matches an action's name, its target and its prompt. Client lets you tick several clients at once and reads No client activity yet until some activity carries a client. Capability appears once the log holds skills to choose from. Initiator lists the people and channels behind the actions. Status offers Success, Failed, Awaiting and Running: Success includes partly finished work, Failed includes cancelled work, and Awaiting means an approval is waiting for a decision.
Date sets the window the tiles and the rows use, from 24 hours to 90 days. The log keeps no more than 90 days, so it is not a permanent archive.
The default view leaves out the automatic wrapper rows that record a playbook run and its steps, because the real outputs of a run, such as post ideas or approvals, appear as their own rows. Tick Playbook runs only to see just the run rows instead. Routine housekeeping runs never appear in this log.
What one action is, and what is inside it
Read across a row for the time, client, conversation, what was asked, who initiated it, the outcome, and, further right, the status, credits and duration. The Initiated by column shows the person, Sydekick for the agent's own scheduled work, Slack with a link back to the Slack thread, or System. A question a client asked in their own portal shows the client's name and an In client portal label; work the agency did on a client's behalf through the portal is labelled as such.
Open a row to see what was recorded. A conversation shows Every ask in order, the Steps it took, and its Outputs, with Open conversation. A request that needs approval shows who requested it, whether it was drafted and queued, who it is waiting on, and its items, with Open in Inbox. A run shows its schedule and trigger and links to the run's page. A row shows only the sections that exist for that activity.
Reading the outcome
The Outcome column is the quickest read of where something stands: for example Sent to Inbox, N awaiting decision, Approved by a person, Rejected, Post created or No asset. An approval row records a person's decision. It does not by itself prove that a post or reply reached the public platform afterwards, so check the post or review itself for its final state. Likewise, a row labelled Post published is recorded when the post is created; its real status (draft, scheduled or live) is on the post.
Export the audit log
Export downloads the rows you have loaded, in the order you see them, as a CSV named with the date. Your filters apply, so a Failed only view exports as exactly that. The columns cover the time, action, target, initiator, status, credits, duration and capability, and recorded steps come too when Show step detail is on. If more rows exist than are loaded, select Load more first. The button is greyed out on an empty table.
Export is manual and per view. There is no scheduled or emailed version of this report.
Who can see what
If you only have access to some clients, you see only the rows for those clients. A person without Audit logs page access does not see the page at all.
What you'll see when it's done
- The rows for the client, status and date range you chose, with their tiles
- An opened row showing its asks, steps and approval state
- A downloaded CSV of what was on screen
Watch out for
- The tiles are sums of the loaded rows. They change when you filter or load more, and they are not the account's credit balance.
- An approval row is a decision, not a delivery. Check the post or the review for its final state.
- There is no read versus write column. Steps list what ran, not whether each step only read data.
Questions people ask
Why are the credits here different from Agent Home?
Because they answer different questions. Home reports the account's credit period from the credit ledger. This page adds up the credits shown on the rows it has loaded, over a calendar-month or rolling window. Use Home for what is left.
Where did my playbook runs go?
Hidden by default. The automatic run and step rows are left out of the default view, and the outputs show as their own rows. Tick Playbook runs only to see the runs.
How far back does the log go?
At most 90 days, with Last 90 days the longest option.
Can I schedule an export?
No. Export is manual and covers the rows on screen.