This page explains how to see what agents did with your vault: the activity log (the audit log, in the command line’s words), with every request, its reason and its outcome; how to check that nobody rewrote it; and the report of what one agent saw. You read it on your devices; an agent has no command or tool for it.
klav log
klav logWHEN WHO WHAT ITEM REASON ALERTS
2026-11-12 23:03:12Z claude-code agent.access totp → allow klav://Personal/General/GitHub Finish signing in to GitHub to open the release PR first_use
2026-11-12 23:03:12Z you approval.decided totp → approve klav://Personal/General/GitHub
2026-11-12 23:03:00Z claude-code agent.access totp → ask klav://Personal/General/GitHub Finish signing in to GitHub to open the release PR
2026-11-12 23:02:57Z claude-code agent.access inject → allow klav://Personal/General/Stripe Write the test config
2026-11-12 23:02:57Z claude-code agent.access inject → allow klav://Personal/General/Stripe Run the billing tests against Stripe's test mode first_use
2026-11-12 23:02:48Z you agent.granted for "claude-code" (5d1c0a7e)
2026-11-12 23:02:37Z claude-code agent.registered
This device has not verified the log yet: run `klav log verify`.Newest first: when (UTC), who (you or the agent’s name), what it did and with what outcome (allow, ask or deny from the policy, approve from you), on which item, and the agent’s reason. A grant you signed says which agent it was for: its name and the start of its ID.
Alerts:
first_use: the first time that agent used that item’s secrets;burst: an agent used a secret after 20 or more requests in the minute before;denied: the policy denied it.
Filters: --item, --agent, --vault, --since and --until (24h, 7d, a date or a UTC time), --alerts, --permission, --outcome and --text. --json and --csv export every event that passes them. klav log --task <id> summarizes one task: who asked and approved it, every use, what was asked outside it, and how it ended.
In the Mac app, the log is Development › Activity (⌘5), and each item has “Who used it”. On the iPhone, it is “Activity” in the Agents sheet.
Check that nobody rewrote it
Each event is chained to the one before by a hash, and agents and your devices sign what they do. klav log verify checks the whole chain:
klav log verifyThe audit log is intact: 9 events, 4 agent signatures, 2 account signatures, 0 signed heads.
This device's signed head is at event 9.It checks every agent’s signature against what the event says, and that the server didn’t shorten the log or change an event this device already saw. Then it signs the new head, unless this device already signed it. If something was changed, removed or reordered, it exits with code 3 and says from where. The first time on a device it also warns that earlier events are trusted as the server sent them. If an agent’s request was logged after you revoked that agent, it warns about each one (“acted after its revocation”) but still exits with 0: an older server could log a request already on its way when you revoked the agent.
What an agent saw
klav agent show <name> shows an agent, its grant and a report built from the log:
klav agent show claude-codeAGENT STATUS PAUSED PROJECTS WORDS ID
claude-code active Personal/General ferocity mothproof bolt cassette gleeful seizing …
Grant v1 (0 rules).
claude-code held the keys of: Personal/General. It could read the titles, usernames and URLs of their items.
What claude-code used (from the audit log):
ID ITEM PERMISSIONS SECRET USES LAST
ba8fb3dd klav://Personal/General/GitHub totp yes 1 2026-11-12 23:03:12Z
d0598530 klav://Personal/General/Stripe inject yes 2 2026-11-12 23:02:57Z
warning: It obtained the secrets marked `yes`: change them where they are used (the site, the API) if you no longer trust it.- The projects whose keys it held: it could read their items’ titles, usernames and addresses.
- Each item it used, with which permissions, how many times and when.
SECRETisyeswhen the agent obtained a value: through a command it ran, a file it wrote, a page it filled, a code, a value it revealed or an attachment delivered to it. “Obtained” means the value reached the agent’s runtime, even if it never reached the agent’s model.- How many of its attempts were denied, if any.
klav agent revoke prints the same report, and the Mac and iPhone show it when you revoke an agent.