catalog / cross-platform

Past-chat search has a different shape on each platform

verified

Searching past conversations uses conversation_search on Claude.ai, a different shape on Desktop, and isn't available on Claude Code — same intent, three implementations.

  • Category: cross-platform
  • Applies to: Claude.ai ✅ (keyword + time) | Claude Code ❌ (file-based memory) | Claude Desktop ✅ (session list, no keyword)
  • Verified on: 2026-05-17
  • Verification tier: verified
  • Severity: Feature gap

Symptom

A SKILL.md that says "find what we discussed about X last week" succeeds, fails, or produces wildly different results depending on the platform.

Cause

Three different mechanisms:

  • Claude.ai: conversation_search (keyword) + recent_chats (time-windowed)
  • Claude Code: no chat search — uses file-based memory (MEMORY.md + memory/*.md)
  • Claude Desktop: mcp__session_info__list_sessions + read_transcript (deferred, no keyword search)

Fix

If a skill needs past-chat search, target one platform explicitly or describe the intent generically and accept divergent behavior. Don't write portable past-chat-search skills.

Notes

Related: ../tool-behaviors/conversation-search-keyword-only-matching.md, ../tool-constraints/recent-chats-max-n.md.