feat: Add memory management tools — view, edit, delete L1 memories

[Enominera]

Problem

Currently, the plugin provides tdai_memory_search and tdai_conversation_search for querying memories, but there is no way to:

1. List all stored memories — users cannot see the full inventory of what the agent has learned
2. Edit a memory — if a memory is outdated or incorrect, there is no way to correct it
3. Delete a memory — if a memory is wrong or no longer relevant, there is no way to remove it

Use Case

In long-term collaborative scenarios, memory accuracy degrades over time:

• User preferences change (e.g., "user prefers dark mode" → later switches to light mode)
• Facts become outdated (e.g., "user works at Company A" → later changes jobs)
• The LLM may extract incorrect memories from ambiguous conversations

Without management tools, users must either:

• Wait for the memory to be naturally superseded (which may never happen)
• Reinstall the plugin and re-seed (losing all memories)
• Manually edit the SQLite database (risky and not user-friendly)

Proposed Solution

Add a set of CLI commands and/or agent tools:

CLI Commands

# List all memories (with filters)
openclaw memory-tdai list [--type persona|episodic|instruction] [--limit 50]

# Show a specific memory
openclaw memory-tdai show <memory_id>

# Edit a memory
openclaw memory-tdai edit <memory_id> --content "updated content"

# Delete a memory
openclaw memory-tdai delete <memory_id>

# Purge memories matching a query
openclaw memory-tdai purge --query "outdated info" [--confirm]

Agent Tools

Add tdai_memory_edit and tdai_memory_delete tools that the agent can use when the user asks to correct or remove a memory.

Current Workaround

I wrote a Python script that reads the L1 JSONL files and displays all memories:

import json
from pathlib import Path

records_dir = Path("~/.openclaw/memory-tdai/records")
for f in sorted(records_dir.glob("*.jsonl")):
    for line in f.read_text().strip().split("\n"):
        if line:
            r = json.loads(line)
            print(f"[{r.get(type)}] {r.get(content)[:200]}")

This works for viewing but cannot edit or delete.

Priority

This is important for trust and adoption. Users need to feel in control of what the agent "remembers" about them.

[Maxwell-Code07]

Thanks for the feature request! Memory management (view/edit/delete) is a valid need for long-term usage. We'll evaluate this internally. Appreciate the detailed proposal and workaround script!