Connect existing run evidence

v0.19.0 adds Agent Trace export and Entire checkpoint import. These are evidence-conversion paths, not replacements for your coding agent or orchestration framework.

Export Agent Trace

Terminal

tine export result.tine --format agent-trace
tine export result.tine --format agent-trace --output trace.json

The export writes one Agent Trace RFC 0.1.0 record, using the schema's version: "0.1" spelling. Successful recorded tool writes become file entries, with AI contributor attribution and recorded provider/model identity. The run ID is retained under metadata["io.github.0xcircuitbreaker.opentine"].

Line ranges are emitted only for a final whole-file write with known coverage. Edits, patches, and shell redirects receive file-level attribution with empty ranges, not guessed line numbers. Paths outside the enclosing repository are omitted and counted. VCS details and conversation URLs are omitted when not recorded. --endpoint is refused for this format.

Import Entire checkpoints

Terminal

# Newest checkpoint, last session
tine import . --format entire --save run.tine
# Chosen checkpoint and zero-based session index; runs is an existing repo
tine import .::a3b2c4d5e6f7/0 --format entire --repo runs

The source is a git repository containing entire/checkpoints/v1, or its origin remote-tracking branch when no local branch exists. The importer reads git objects without checking out the branch. REPO::ID[/N] selects a checkpoint by ID or unique prefix and optionally a zero-based session index. Auto format detection recognizes a directory source.

Claude Code, Codex, and Gemini CLI transcripts reuse their existing importers; other agents import the prompt. Checkpoint IDs, agent, model, touched files, and Entire token usage are preserved under attributes["entire"]. The supported layout was read from entireio/cli commit 54b6a5c; upstream format changes may require updates.

Evidence boundaries

Blob sizes are checked before reading and session paths must match the sharded checkpoint layout. Imported sessions remain source evidence, not independent attestation. Agent Trace output describes recorded writes, not proof of every process action. Inspect sensitive prompts and file paths before sharing either output.

See import destinations and pricing, other export formats, and the upgrade guide.