Currently, tool results are only pruned from context via automatic TTL (e.g., 1h with cache-ttl mode). This creates inefficiency when large tool results (emails, search results, etc.) are only needed momentarily.

Use case:

1. User asks: "Find my first iPhone purchase in Gmail"
2. Agent fetches 25 emails (80k+ tokens of results)
3. Agent answers the question immediately
4. Those 80k tokens sit in context for the next hour despite being useless

Requested feature:
A way to manually purge tool results from context once they're no longer needed.

Possible implementations:

• CLI command: openclaw context clear --tool-result <id>
• Agent-accessible function: context_clear(tool_result_id)
• Auto-clear flag on tool calls: {clearAfterUse: true}

Benefits:

• Reduces context bloat
• Extends session lifespan before compaction
• Improves performance for search-heavy workflows

Current workaround: Wait for TTL expiration (wasteful)