archived-session-management（存档会话管理）

Archived Session Management

Implement archived-session UX as an end-to-end feature, not a single-button patch. Archived session work spans the Control UI, gateway RPC handlers, transcript file moves, the session_history SQLite index, and the live sessions.json store.

Core workflow

1. Inspect the whole chain first

- UI list/search rendering - gateway methods for sessions.archived, sessions.resume, sessions.deleteArchived - archive/restore helpers in src/gateway/session-archive.ts - history DB filters in src/config/sessions/history-db.ts - live session store behavior in sessions.json

1. Treat these as separate failure classes

- archived list cannot find a real session - archived list finds the session but resume fails - resume works but opens the wrong live session - session vanishes from both Live and Archived due to DB/store drift

1. Fix backend truth before UI polish

- Make archived search and restore semantics correct. - Then wire or refine UI buttons, copy, filters, and pagination.

1. Reconcile legacy/inconsistent rows when needed

- Archived session features often fail because the DB, transcript path, and live store disagree. - Be tolerant in reads, then repair writes so future rows stay consistent.

Required implementation areas

1. Archived-session search

Archived search must match more than just friendly labels. Include at least:

• sessionId
• sessionKey
• displayName
• firstMessage
• agentId
• channel
• chatType
• status
• filePath

When filtering archived sessions, do not rely on status = 'archived' alone. Legacy rows can still be archived-discoverable when:

• archivedAt IS NOT NULL, or
• filePath points into /sessions/archive/

Read references/implementation-map.md before editing the query layer.

2. Restore / resume behavior

Restoring an archived session must be idempotent.

Handle all three cases:

• transcript exists only in archive → move it back to active sessions dir
• transcript already exists in active sessions dir → do not fail; repair metadata and continue
• transcript exists nowhere → return not found

When restore succeeds, update the history row so it reflects the active transcript path and active status.

3. Rebind restored sessions into a unique live key

Do not resume old archived main sessions into agent:main:main.

Older rows often store that key, which causes the UI to jump back into the current main thread instead of the restored transcript.

Bind restored transcripts to a distinct live key such as:

• agent:<agentId>:archive-<sessionId>

Then write that binding into the live session store so the session actually appears in Live and can be opened by URL.

4. UI expectations

Archived-session UI should include:

• Archived tab / section
• search box
• Resume button
• Delete action for archived transcripts when supported
• loading / empty states that are specific enough to debug

If the repo already has the UI shell, wire it to the corrected backend instead of rebuilding it.

Debugging checklist

When a user says an archived session is broken, verify all four:

1. History row exists in session_history
2. Transcript file exists at the path the row expects, or at the active fallback path
3. History row status/path fields are coherent
4. Live store contains a session binding after resume

If a session is in neither Live nor Archived, suspect DB/store drift immediately.

File map

Read references/implementation-map.md for the concrete OpenClaw file map, bug patterns, and example fixes.

Delivery standard

When work is complete, report:

• what failed
• which layer was wrong (UI, restore logic, DB filter, store binding, or mixed state)
• what changed
• whether a rebuild / gateway restart is needed