Debugging with Multiple Agents

# Launch investigators
hcom 1 claude --tag debug-db --go --headless \
  --hcom-system-prompt "You investigate database-related issues. \
  Check: queries, indexes, connections, migrations, data integrity."

hcom 1 claude --tag debug-api --go --headless \
  --hcom-system-prompt "You investigate API-related issues. \
  Check: endpoints, middleware, validation, serialization, auth."

hcom 1 claude --tag debug-infra --go --headless \
  --hcom-system-prompt "You investigate infrastructure issues. \
  Check: containers, services, networks, resources, configs."

# Launch coordinator
hcom 1 claude --tag debug-lead --go --hcom-prompt "
Bug report: Users seeing 500 errors on /api/users endpoint.
Error rate: 30% of requests failing.

Distribute investigation:

1. Send to specialists:
   hcom send '@debug-db Check database for /api/users issues' --thread debug-123
   hcom send '@debug-api Check API code for /api/users issues' --thread debug-123  
   hcom send '@debug-infra Check infra for /api/users issues' --thread debug-123

2. Wait for findings (timeout 180s):
   hcom events --wait 180 --thread debug-123 --type message

3. Analyze findings:
   - Identify root cause
   - Propose fix
   - Send summary to @reporter
"

hcom send '@debug-lead 500 errors on /api/users, 30% failure rate' \
  --from reporter --thread debug-123

hcom 1 claude --tag debugger --go --hcom-prompt "
Issue: Login form not submitting

Systematic debugging:

1. Check browser console:
   - Any JavaScript errors?
   - Network requests failing?
   - Broadcast: hcom send '@team BROWSER: [findings]'

2. Check API logs:
   - Is /api/auth/login receiving requests?
   - What status codes?
   - Broadcast: hcom send '@team API: [findings]'

3. Check database:
   - Are queries executing?
   - Any connection issues?
   - Broadcast: hcom send '@team DB: [findings]'

4. Check auth service:
   - Is JWT validation working?
   - Token generation?
   - Broadcast: hcom send '@team AUTH: [findings]'

5. After each check:
   - If issue found: STOP and report
   - If no issue: Continue to next
"

hcom 1 claude --tag debugger --go --headless \
  --hcom-prompt "Debug the pagination bug in src/api/users.ts. \
  Investigate thoroughly. When you find the issue, send findings to @reviewer."

hcom 1 claude --tag reviewer --go --hcom-prompt "
Wait for @debugger's findings:
  hcom listen --from debugger --timeout 300

When received:
1. Review their analysis
2. Check their evidence
3. Try to reproduce their conclusion
4. Verify the fix they propose

Send verdict:
  hcom send '@debugger @lead [APPROVED/NEEDS_MORE_INVESTIGATION]: [reasoning]'
"

#!/usr/bin/env bash
# Parallel git bisect
set -euo pipefail

commits=($(git log --oneline -n 20 | cut -d' ' -f1))
mid=$((${#commits[@]} / 2))

# Test first half
hcom 1 claude --tag bisect-a --go --headless --hcom-prompt "
git checkout ${commits[0]}
npm test -- users.test.ts
if passing: hcom send '@lead FIRST_HALF: passing'
if failing: hcom send '@lead FIRST_HALF: failing'"

# Test second half  
hcom 1 claude --tag bisect-b --go --headless --hcom-prompt "
git checkout ${commits[$mid]}
npm test -- users.test.ts
if passing: hcom send '@lead SECOND_HALF: passing'
if failing: hcom send '@lead SECOND_HALF: failing'"

# Coordinator narrows down based on results
# Continue bisecting in failing half

hcom 1 claude --tag reproducer --go --headless \
  --hcom-prompt "Reproduce the bug: \
  1. Visit http://localhost:3000/login \
  2. Enter email: [email protected], password: wrong \
  3. Submit form \
  4. Observe behavior \
  \
  Send observations to @monitor"

hcom 1 claude --tag monitor --go --hcom-prompt "
Monitor while @reproducer tests:

1. Tail logs:
   tail -f logs/api.log logs/db.log

2. Watch for errors/warnings

3. Check metrics:
   - Response times
   - Error rates
   - Resource usage

4. When @reproducer sends observations:
   - Correlate with log timestamps
   - Identify anomalies
   - Send diagnosis to @reproducer @lead
"

#!/usr/bin/env bash
# Multi-agent debugging workflow
set -euo pipefail

issue="$1"
if [[ -z "$issue" ]]; then
  echo "Usage: hcom run debug 'description'"
  exit 1
fi

thread="debug-$(date +%s)"
LAUNCHED_NAMES=()

cleanup() {
  for name in "${LAUNCHED_NAMES[@]}"; do
    hcom stop "$name" --go 2>/dev/null || true
  done
}
trap cleanup ERR

track_launch() {
  names=$(echo "$1" | grep '^Names: ' | sed 's/^Names: //')
  for n in $names; do LAUNCHED_NAMES+=("$n"); done
}

echo "Starting debug session: ${issue}"
echo "Thread: ${thread}"

# Launch specialists
echo "Launching specialists..."

out=$(hcom 1 claude --tag debug-code --go --headless \
  --hcom-system-prompt "Code debugger. Check: logic errors, type errors, null checks, edge cases." 2>&1)
track_launch "$out"

out=$(hcom 1 claude --tag debug-data --go --headless \
  --hcom-system-prompt "Data debugger. Check: DB queries, data flow, state management, caching." 2>&1)
track_launch "$out"

out=$(hcom 1 claude --tag debug-env --go --headless \
  --hcom-system-prompt "Environment debugger. Check: configs, env vars, dependencies, services." 2>&1)
track_launch "$out"

# Launch coordinator
out=$(hcom 1 claude --tag debug-lead --go --hcom-prompt "
ISSUE: ${issue}
THREAD: ${thread}

Debug workflow:

1. Distribute investigation:
   hcom send '@debug-code @debug-data @debug-env Investigate: ${issue}' \
     --thread ${thread} --intent request

2. Collect findings (wait 180s):
   hcom events --wait 180 --thread ${thread} --type message --last 10

3. Analyze findings:
   - Look for common threads
   - Identify most likely root cause
   - Verify with targeted tests

4. Propose fix:
   - What to change
   - Why it fixes the issue  
   - How to verify

5. Send summary:
   hcom send '@team ## Debug Summary
   
   ROOT CAUSE: [description]
   FIX: [description]
   VERIFICATION: [steps]' --thread ${thread}

6. Stop specialists:
   hcom stop @debug-code @debug-data @debug-env
" 2>&1)
track_launch "$out"

echo
echo "Debug session active. Watch progress:"
echo "  hcom events --wait 300 --thread ${thread}"
echo
echo "Or in TUI:"
echo "  hcom"
echo

# Wait for completion
hcom events --wait 300 --thread "${thread}" \
  --sql "msg_text LIKE '%Debug Summary%'" >/dev/null || true

echo
echo "Debug session complete. Results:"
hcom events --thread "${thread}" --type message --last 5

# Cleanup
cleanup
trap - ERR

hcom run debug "Login form returns 500 on submit"
hcom run debug "Pagination broken after deploy"
hcom run debug "Memory leak in API server"

# Agent subscribes to file edit events
hcom events sub --file '*.py' --context 'tool:Edit'

# When file changes, agent re-runs tests

# Agent watches for test command execution
hcom events sub --cmd 'pytest'

# When tests run, agent checks results and reports failures

# Watch what debugger is doing
hcom events --agent debugger --last 10 --wait 60

# Coordinator reconciles:
hcom send '@debug-code @debug-data Which finding explains ALL symptoms?' \
  --thread debug-123

# Query what agents found so far:
hcom events --thread debug-123 --type message --last 20

# Extend investigation:
hcom send '@debug-lead Continue investigation, focus on [area]' --thread debug-123

# Verify with second agent:
hcom send '@reviewer Verify debugger's finding: [description]' --thread debug-123