Agent execution logs are daemon-owned audit material, not Mission artifacts by default. They are retained with daemon runtime state, and with the Mission record/state store when the execution is Mission-scoped, so the system can inspect delivered interaction, recover context, and preserve evidence of what was sent to and produced by an Agent adapter.

Agent execution logs are not one undifferentiated transcript. Mission distinguishes at least two log families:

• AgentExecution interaction journals: semantic execution truth used for replay, audit, observations, decisions, state effects, and UI projections.
• AgentExecution terminal recordings: raw PTY transport audit for terminal input, output, resize, and exit records.

Mission workflow event logs remain separate orchestration truth and must not become AgentExecution transcripts.

Mission control should not show every raw session transcript as a Mission artifact. This keeps the operator outline focused on curated Mission work rather than terminal exhaust.

When log-derived material becomes useful to the Mission, the daemon or operator promotes a curated derivative into a separate Agent-session artifact. Examples include a transcript summary, extracted test output, patch summary, or generated implementation note. The promoted artifact may reference or quote the raw Agent execution log, but it is not the raw log itself.

This keeps runtime interaction durable and inspectable while preserving a small, intentional Mission artifact model. ADR-0006.08 defines the semantic interaction journal as the canonical replayable AgentExecution log. Terminal recordings remain raw transport material.