CSV audit export omits origin, record_kind, and metadata columns

[olivrg]

Summary

The CSV audit export omits the origin, record_kind, and metadata columns, even though they exist in the audit schema and are returned by the JSON export. This makes CSV exports inconsistent with JSON, and means a CSV export filtered by one of these fields has no column showing the value it was filtered on.

Background

origin / record_kind / metadata were added to the audit schema in #12 and are surfaced throughout the dashboard in #16 (Origin column, record-kind chips, channel/sender columns + filters). #16 also made /api/audit/export accept origin, record_kind, channel_id, and sender_id filter params. But the CSV column list was never updated to include the new fields.

Gap originates in #12 (columns added to the schema without CSV coverage); surfaced by the export filters added in #16. Tracked separately as it predates #16 and is out of its scope.

Current behaviour

• JSON export (GET /api/audit/export?format=json) returns the full AuditRecord[], so origin/record_kind/metadata are present.
• CSV export (?format=csv, used by both the dashboard API and the CLI export) is driven by a hardcoded header list that lacks all three: packages/proxy/src/audit/csv.ts — CSV_HEADERS (lines 8–33).
• Consequence: GET /api/audit/export?format=csv&origin=openclaw correctly returns only OpenClaw rows, but the CSV has no origin column to show it — filterable but not visible. Same for record_kind and the metadata.channel_id / metadata.sender_id context.

Expected behaviour

CSV export should include origin, record_kind, and metadata so it matches the JSON export and the dashboard's column set.

Proposed fix

• Add 'record_kind', 'origin', and 'metadata' to CSV_HEADERS in packages/proxy/src/audit/csv.ts. recordToRow (lines 64–74) already JSON-stringifies object-typed values, so metadata serializes with no further change (and the existing formula-injection guard in csvEscape still applies).
• Update packages/proxy/src/audit/csv.test.ts to assert the new columns.
• Update the "CSV Format" section of docs/audit.md (note metadata serializes as a JSON string, like tool_input / evidence_chain).

Notes / decisions for the implementer

• Column ordering: appending the three new headers at the end of CSV_HEADERS is the lowest-churn option and keeps existing column positions stable for any downstream consumers parsing by index.
• This is purely additive to the export surface; no schema or API-contract change.

Related

• Gap introduced by Sideband decision API: framework-neutral governance endpoints for hook-based adapters #12 (audit schema columns).
• Surfaced by Dashboard: render adapter-origin tool calls #16 (export filters + dashboard rendering).