Module export 

Trace export (Task 12): metadata export of persisted synaps-request-trace/1 records, explicit redacted content export, and the recursive secret redaction they share.

• Metadata export (synaps trace export <id> --metadata-only): reads the private trace JSONL, validates every line as a RequestTrace, selects the records whose turn_id or request_id equals the given (validated, bounded) TraceId, and writes them to a user-selected private file (0600, parent 0700), refusing symlinks and pre-existing targets. No network, no content — the records are structurally metadata-only.
• Content export (--include-content --allow-content-export): fail-closed. It requires BOTH flags in the same invocation and an existing ephemeral capture bundle produced by /trace next content (see trace::controls). The bundle is re-redacted (defense in depth), written under its own clearly-labeled schema (synaps-trace-content-export/1, never synaps-request-trace/1), and the capture bundle is deleted (consumed). Expired bundles are deleted and refused.

Structs

ContentExport

On-disk shape of the content export artifact. NOT a request trace.

MetadataExportStats

Result of a metadata export.

Enums

ExportError

Typed export failure. Every variant is metadata-only (paths, counts).

Constants

CONTENT_EXPORT_SCHEMA

Schema tag of the content-export artifact. Deliberately distinct from both the request-trace and capture schemas.

Functions

export_content

Export the ephemeral capture bundle for id into a fresh private file. Fail-closed: requires allow_content_export, an existing unexpired bundle, and re-redacts before writing. Consumes (deletes) the bundle.

export_metadata

Export all records whose turn_id or request_id equals id from the trace JSONL at trace_log into a fresh private file at output.

redact_command_text

AGGRESSIVE scrub for one-off display of a command the user is about to approve (the Confirm prompt preview and the confirmer’s message). It trades false positives for recall, so it must NOT be used for stored text the model or user later reads back (context archive, memory, traces): those use redact_value / redact_text only. On top of redact_text it adds:

redact_text

Free-text scrub through the shared, high-precision rules ([scrub_secret_patterns] plus the word-level known-prefix scrub). Returns the input unchanged when clean.

redact_value

Recursively redact credential-bearing keys and obvious secret-shaped values in place. Returns the number of redactions applied. Applies to nested objects and arrays at any depth.

sweep_expired_captures

Delete any expired capture bundles in dir (bounded retention sweep).