discord-clawd（Discord clawd 搜索）

Discord Clawd

Use this for Discord questions first. Local archive first; live Discord only when freshness matters and the archive is stale.

Trigger

Use when the user asks things like:

• "what happened in #maintainers"
• "who posted the most in discord"
• "search discord for..."
• "how many messages/users/channels"
• "summarize this week in #channel"
• "is Discord fully synced"

Data Sources

Prefer in this order:

1. ~/.discrawl/discrawl.db
2. /tmp/discrawl --config ~/.discrawl/config.toml...
3. Live refresh from ~/Projects/discrawl only if needed

Do not browse the web for Discord archive questions unless the user explicitly wants external context too.

Freshness Check

For latest, recent, today, this week, or sync questions, check freshness first.

Useful queries:

sqlite3 ~/.discrawl/discrawl.db \
  "select coalesce(max(updated_at),'') from sync_state where scope like 'channel:%';"

sqlite3 ~/.discrawl/discrawl.db "
select count(*)
from channels c
where coalesce(c.kind,'') in (
  'text','news','announcement',
  'thread_public','thread_private','thread_news','thread_announcement'
)
and not exists (
  select 1 from sync_state s
  where s.scope = 'channel:' || c.id || ':history_complete'
)
and not exists (
  select 1 from sync_state s
  where s.scope = 'channel:' || c.id || ':unavailable'
);"

Interpretation:

• 0 remaining = fully synced for real message-bearing channels
• forum parents are metadata containers; their post history lives in thread channels

Query Workflow

1. Resolve scope: channel, date range, author, keyword.
2. Check freshness if the user asked for recent/current data.
3. Use SQL for counts/rankings/exact filters.
4. Use search/FTS for discovery, then SQL for precise slices.
5. Prefer the messages CLI for exact channel+date slices before dropping to raw SQL.
6. Report with absolute dates, counts, and channel names.

Common Queries

Resolve channel name:

sqlite3 ~/.discrawl/discrawl.db \
  "select id, name, kind from channels where name like '%maintainers%' order by name;"

Top channels:

sqlite3 ~/.discrawl/discrawl.db "
select c.name, c.kind, count(m.id) as messages
from channels c
join messages m on m.channel_id = c.id
group by c.id
order by messages desc
limit 20;"

Top authors with names:

sqlite3 ~/.discrawl/discrawl.db "
select
  coalesce(mem.display_name, mem.username, m.author_id) as author,
  m.author_id,
  count(*) as messages
from messages m
left join members mem
  on mem.user_id = m.author_id
 and mem.guild_id = m.guild_id
where coalesce(m.author_id,'') != ''
group by m.author_id, coalesce(mem.display_name, mem.username, m.author_id)
order by messages desc
limit 20;"

Messages in one channel for the last X days:

/tmp/discrawl --config ~/.discrawl/config.toml \
  messages --channel maintainers --days 7 --all

Messages in one channel since a date:

/tmp/discrawl --config ~/.discrawl/config.toml \
  messages --channel '#maintainers' --since 2026-03-01T00:00:00Z --all

Keyword search:

/tmp/discrawl --config ~/.discrawl/config.toml search "query terms"

Read-only SQL via CLI:

/tmp/discrawl --json --config ~/.discrawl/config.toml sql \
  "select count(*) from messages;"

Summaries

For channel summaries:

• pull the exact slice first
• include counts: messages, authors, date span
• identify recurring topics, decisions, blockers
• do not claim certainty beyond the retrieved slice

For release-window summaries, anchor on exact timestamps, not vague relative dates.

Name Resolution

Prefer:

1. members.display_name
2. members.username
3. messages.raw_json -> $.author.username
4. raw author id

If a user asks "who is this id", resolve locally before answering.

Live Refresh

Only do this if the user wants fresher data than the archive has.

Repo:

cd ~/Projects/discrawl

Token source:

ssh peters-mac-studio-1 \
  "zsh -lc 'source ~/.profile >/dev/null 2>&1; printf %s \"\$DISCORD_BOT_ID_FLAWD\"'"

Targeted refresh is better than full sync for ad hoc questions:

DISCORD_BOT_TOKEN="..." /tmp/discrawl \
  --config ~/.discrawl/config.toml \
  sync --full --channels <channel-id> --concurrency 1

Guardrails

• never expose bot tokens
• prefer local DB over network
• use absolute dates in answers
• if freshness is uncertain, say so
• if the archive is stale, either sync first or label the answer as archive-based