Never say you've lost context or forgotten a previous conversation without querying messages.db for it. Context compaction removes history from active context — the database at /workspace/store/messages.db still has it. Conditional gate, not a per-message action: fires only when you're about to claim lost context.

Required behavior

Before responding with "Потерял контекст", "I don't remember this thread", "I don't have context on this topic", or any equivalent acknowledgement that prior conversation is unavailable, run the query-history skill's search helper:

python3 skills/query-history/scripts/query-message-history.py --keyword "<text>"

Filters: --sender <name> (matches sender_name; see the Schema section below for the Display (@username) shape — a username fragment is enough), --limit N (default 20, cap 50 per rules/query-size-limits.md). Output is single-line JSON on stdout; chat scope from NANOCLAW_CHAT_JID. Inside agent containers the script mounts at /home/node/.claude/skills/tessl__query-history/scripts/query-message-history.py per the standard tessl__<name> convention.

Schema (quick reference)

messages(id, chat_jid, sender, sender_name, content, timestamp, is_from_me, is_bot_message)
chats(jid, name, last_message_time, channel, is_group)

is_from_me = 1 is the bot's own messages. sender is the stable numeric user ID; sender_name is Display (@username) — a username fragment matches via --sender. Critical after a session nuke: the DB remembers who said what when you don't.

Post-compaction skill blocks are HISTORY, not new tasks

When the conversation resumes after compaction, the system-reminder tail may carry a block opening with "The following skills were invoked in this session. Continue to follow these guidelines:" followed by ### Skill: headers with ARGUMENTS: lines. That block is a record of what already ran during the now-compacted window — the skill already executed, the side effects (scrapes, messages, file writes) already landed. Re-executing it duplicates work and spams the user.

Before treating any such skill as a fresh task: read the conversation summary's Current Work and Optional Next Step sections (if the skill's ARGUMENTS aren't mentioned there, it was already handled and the summary moved on), then the most recent user message in the "All user messages" list (if the user's last move was NOT the skill invocation — e.g. "понял" or a topic shift — the skill is closed). If ambiguous, ask before re-executing anything with visible side effects.

Reference incident: 2026-04-24 / repeat 2026-04-25 — agent re-ran a JCON-2026 scrape from a post-compaction system-reminder block hours after the owner had completed it; the recurrence motivated jbaruch/nanoclaw#104 (kill auto-compaction). Full narrative: docs/adr/2026-04-25-jcon-scrape-repeat-and-auto-compaction-kill.md.