Search Messages

Search Telegram Messages

Search messages globally across all chats or within a specific chat.

Self-Evolving Skill: This skill improves through use. If instructions are wrong, parameters drifted, or a workaround was needed — fix this file immediately, don't defer. Only update for real, reproducible issues.

Preflight

1. Session must exist: ~/.local/share/gramjs/<profile>.session
   • If missing, run /tlg:setup first

Usage

/usr/bin/env bash << 'EOF'
ROOT="$(cc-plugin-root tlg)"
SCRIPT="$ROOT/scripts/tg-cli.ts"

# Global search (all chats)
bun "$SCRIPT" search "search term" -n 20

# Search in specific chat
bun "$SCRIPT" search "keyword" --chat 2124832490

# Filter by sender
bun "$SCRIPT" search "topic" --from @username

# Combined: search in chat from specific sender
bun "$SCRIPT" search "query" --chat @groupname --from @sender -n 10
EOF

Parameters

Output Format

Default (full text, multi-line indented):

[YYYY-MM-DD HH:MM] [Chat Name] (id:12345) Sender: First line of body
    continuation line 1 (indented under header)
    continuation line 2

With --preview 100:

[YYYY-MM-DD HH:MM] [Chat Name] (id:12345) Sender: First line of body ⏎ continuation collapsed onto…

Use --preview N only for high-volume scans where you want one row per message. Default = full text — long bodies are no longer silently truncated at 150 chars (the prior default forced manual workarounds when content actually mattered).

Anti-Patterns

• Flood risk: Global search with common terms may hit rate limits (~30s wait per 10 requests)
• Empty results: Global search requires non-empty query string

Post-Execution Reflection

After this skill completes, check before closing:

1. Did the command succeed? — If not, fix the instruction or error table that caused the failure.
2. Did parameters or output change? — If tg-cli.ts's interface drifted, update Usage examples and Parameters table to match.
3. Was a workaround needed? — If you had to improvise (different flags, extra steps), update this SKILL.md so the next invocation doesn't need the same workaround.

Only update if the issue is real and reproducible — not speculative.