Kiro sessions since 5 Sep are not exported: kiro-cli moved to ~/.kiro/sessions/cli

[NetDevAutomate]

What is happening

Kiro sessions after 5 September never reach the session database, and nothing reports it.

• kiro_cli's newest exported message is 2026-09-05 12:28 UTC. Codex, Claude Code and pi exported normally after that date.
• The Kiro exporter (packages/agent-session-tools/src/agent_session_tools/exporters/kiro.py) reads only ~/Library/Application Support/kiro-cli/data.sqlite3 (tables conversations and conversations_v2). The newest conversations_v2.updated_at is 2026-09-05 12:29 UTC: kiro-cli (now 2.24.0) has written no rows there since.
• kiro-cli keeps its sessions in ~/.kiro/sessions/cli/, one <uuid>.jsonl and one <uuid>.json per session: 20,019 .jsonl files (2 May to today), 11,169 of them non-empty and newer than 5 Sep 13:00. Nothing in agent_session_tools reads that directory.
• The periodic sweep (com.sessionweaver.export-sweep, every 4 h) exits 0 with added: 0, updated: 0, skipped: 893, so the loss is silent.
• The doctor's export_freshness row did not show it either: it reported the newest message across all harnesses (48 h, from Codex).

The new store (key names only)

• <uuid>.json metadata: created_at, cwd, session_created_reason, session_id, session_state, title, updated_at.
• <uuid>.jsonl records: {version, kind, data}; kinds seen: Prompt, AssistantMessage, ToolResults.
• 23 hex-named directories under ~/.kiro/sessions/ are not inspected yet. Open question: which store StudyLoop's web (ACP) Kiro sessions write to.

Decision needed before building

Of the 12,217 sessions created since 5 Sep, 12,212 have session_created_reason: subagent. Most are agent automation: 854 have cwd /, 690 are in ~/.kiro/crew/workspace, and 104 are in the StudyLoop repo. Exporting every session would fill the database with automation transcripts. Which sessions count as the learner's, for example: non-subagent sessions only, or subagent sessions whose cwd is a project?

Recovery

The native files are still on disk, so once the exporter reads this store a backfill recovers the whole gap. Nothing is lost yet.

Done when

• The Kiro exporter reads ~/.kiro/sessions/cli/ under the rule decided above, with fixtures captured from real kiro-cli 2.24.0 files.
• A backfill brings the missing sessions in.
• The doctor's Kiro-store warning (added in the header/doctor clean-up PR) goes quiet on this machine.