Module transcript 

JSONL-based transcript persistence for sub-agent conversations.

Each sub-agent session writes a <task_id>.jsonl file of TranscriptEntry lines and a companion <task_id>.meta.json sidecar with TranscriptMeta.

Files are created with 0o600 permissions on Unix to prevent other users from reading conversation history.

The sweep_old_transcripts function prunes the oldest .jsonl files when a configurable maximum count is exceeded.

Structs

TranscriptEntry

A single entry in a JSONL transcript file.

TranscriptMeta

Sidecar metadata for a transcript, written as <agent_id>.meta.json.

TranscriptReader

Reads and reconstructs message history from JSONL transcript files.

TranscriptWriter

Constants

CHAIN_DOMAIN

Domain-separation tag for this subsystem’s hash chain (issue #6360) — distinct from zeph-session’s so a chain from one subsystem can never verify against the other, and folded into every genesis hash via zeph_common::hash_chain::genesis.

Functions

configure_anchor_store

Configure (or disable, with None) the vault-anchor store for every TranscriptWriter/ TranscriptReader operation in this process from this point forward. See configure_history_integrity’s doc for the single-set-at-startup contract this mirrors.

configure_history_integrity

Configure (or disable, with None) history-chain verification for every TranscriptWriter/TranscriptReader operation in this process from this point forward.

sweep_old_transcripts

Delete the oldest .jsonl files in dir when the count exceeds max_files, plus each deleted file’s companion .meta.json sidecar.