Paste agent event samples above, then select Generate coverage card to see which audit fields are missing, with critical fields like actor identity, session id, approvals, and kill-switch evidence listed first.
About the agent telemetry coverage card
The agent telemetry coverage card turns a pasted set of AI agent event samples into a short report of which audit fields your telemetry is missing. It accepts a JSON array, a single JSON object, or JSONL with one event per line, flattens the keys of every event, and checks them against a fixed checklist of audit fields: actor identity, session or trace id, timestamp, tool call name, tool arguments, approval or authorization decision, data access, policy or guardrail denial, error or exception capture, kill-switch or termination evidence, model or agent version, and outcome. A field is reported missing when no event carries a key that matches it, and the missing fields most worth adding before a production rollout are listed first.
Paste your agent events, then select Generate coverage card. A small synthetic trace is loaded so you can see missing critical fields right away. Everything runs in your browser. The pasted events, their keys, and their values are never uploaded or stored. Download a coverage CSV for the full field-by-field report or copy a markdown review card to paste into a pull request, a design review, or an audit ticket.
How to use
- Paste your agent event samples into the box. A small synthetic trace is loaded so you can see how it works.
- Use a JSON array of events, a single JSON object, or JSONL with one JSON event per line.
- Select Generate coverage card to flatten every event and check it against the audit-field checklist.
- Read the summary and the coverage table. Missing fields are listed first, critical fields before high and medium.
- Select Download coverage CSV for the full report, or Copy review card to paste a markdown summary into a pull request or audit ticket.
Worked examples
A trace with no actor id flags actor identity as critical missing
When none of the pasted events carry a user, actor, or principal key, actor identity is reported missing and listed as a critical gap.
An approval recorded on one event marks approvals present
An event with an approval object, for example approval.approved, satisfies the approval and authorization field even if other events omit it.
camelCase keys are matched too
A key such as sessionId satisfies the session or trace id field, because matching ignores separators and case, so sessionId and session_id are treated the same.
Frequently asked questions
- What does the agent telemetry coverage card check for?
- It checks whether your agent event samples carry the audit fields a reviewer wants before a production rollout. It parses the pasted JSON or JSONL into events, flattens the keys of each event, and matches them against a fixed checklist: actor identity, session or trace id, timestamp, tool call name, tool arguments, approval or authorization, data access, policy or guardrail denial, error or exception, kill-switch or termination, model or agent version, and outcome. A field is reported missing when no event carries a matching key, and critical fields are listed first.
- How does it decide a field is present?
- It matches each field against a small set of candidate key names at whole key-segment boundaries, ignoring separators and case, so sessionId and session_id both satisfy the session field while metadata does not satisfy the data field. It looks only at keys, never at values, so a key that exists with a redaction marker still counts as present. It is a review aid that surfaces likely gaps; it does not confirm that a present field is correctly populated on every event.
- Which input formats does it accept?
- Three: a JSON array of event objects, a single JSON object treated as one event, or JSONL with one JSON value per line. Lines or array elements that are not usable JSON event objects are counted and ignored so a few malformed rows do not block the rest. Large pastes are bounded: input is capped at 1 MB, the event count is capped, and analysis stops early if it would otherwise make the page unresponsive, returning a partial result you can narrow and rerun.
- Are my agent events uploaded anywhere?
- No. Parsing, key flattening, classification, and export all run in your browser. The pasted events, their keys, and their values are never sent to a server or saved. Analytics records only coarse count bands, never the event content, keys, values, or the evidence key paths. Download the coverage CSV or copy the review card before you close the tab.
- Can I keep a record of the review?
- Yes. Download coverage CSV gives you a spreadsheet-ready file with one row per audit field, and any cell that could be read as a spreadsheet formula is escaped so the export is safe to open. Copy review card gives you a markdown summary, with the missing fields first, that you can paste into a pull request, a design review, or an audit ticket.
Use this again tomorrow
Save this page so it's one tap away when you need a quick result.
Ready for a quick Daily Challenge?
Play Daily Challenge on sts.games