trace-claude-code
Automatically trace Claude Code conversations to Braintrust for observability. Captures sessions, conversation turns, and tool calls as hierarchical traces.
System health check (MOT) for skills, agents, hooks, and memory
$ npx -y skills add parcadei/Continuous-Claude-v3 --skill mot --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
/motContext preview
The summary Claude sees to decide when to auto-load this skill.
System health check (MOT) for skills, agents, hooks, and memory
name: mot description: System health check (MOT) for skills, agents, hooks, and memory model: sonnet allowed-tools: [Read, Bash, Glob, Grep]
Run comprehensive health checks on all Claude Code components.
/mot # Full audit (all categories) /mot skills # Just skills /mot agents # Just agents /mot hooks # Just hooks /mot memory # Just memory system /mot --fix # Auto-fix simple issues /mot --quick # P0 checks only (fast)
# Count skills
echo "=== SKILLS ==="
SKILL_COUNT=$(find .claude/skills -name "SKILL.md" | wc -l | xargs)
echo "Found $SKILL_COUNT skill files"
# Check frontmatter parsing
FAIL=0
for skill in $(find .claude/skills -name "SKILL.md"); do
if ! head -1 "$skill" | grep -q "^---$"; then
echo "FAIL: No frontmatter: $skill"
FAIL=$((FAIL+1))
fi
done
echo "Frontmatter: $((SKILL_COUNT - FAIL)) pass, $FAIL fail"
# Check name matches directory
FAIL=0
for skill in $(find .claude/skills -name "SKILL.md"); do
dir=$(basename $(dirname "$skill"))
name=$(grep "^name:" "$skill" 2>/dev/null | head -1 | cut -d: -f2 | xargs)
if [ -n "$name" ] && [ "$dir" != "$name" ]; then
echo "FAIL: Name mismatch $dir vs $name"
FAIL=$((FAIL+1))
fi
done
echo "Name consistency: $((SKILL_COUNT - FAIL)) pass, $FAIL fail"echo "=== AGENTS ==="
AGENT_COUNT=$(ls .claude/agents/*.md 2>/dev/null | wc -l | xargs)
echo "Found $AGENT_COUNT agent files"
# Check required fields
FAIL=0
for agent in .claude/agents/*.md; do
[ -f "$agent" ] || continue
# Check name field exists
if ! grep -q "^name:" "$agent"; then
echo "FAIL: Missing name: $agent"
FAIL=$((FAIL+1))
continue
fi
# Check model is valid
model=$(grep "^model:" "$agent" | head -1 | cut -d: -f2 | xargs)
case "$model" in
opus|sonnet|haiku) ;;
*) echo "FAIL: Invalid model '$model': $agent"; FAIL=$((FAIL+1)) ;;
esac
done
echo "Agent validation: $((AGENT_COUNT - FAIL)) pass, $FAIL fail"
# Check for dangling references (agents that reference non-existent agents)
echo "Checking agent cross-references..."
for agent in .claude/agents/*.md; do
[ -f "$agent" ] || continue
# Find subagent_type references
refs=$(grep -oE 'subagent_type[=:]["'\'']*([a-z-]+)' "$agent" 2>/dev/null | sed 's/.*["'\'']//' | sed 's/["'\'']$//')
for ref in $refs; do
if [ ! -f ".claude/agents/$ref.md" ]; then
echo "WARN: $agent references non-existent agent: $ref"
fi
done
doneecho "=== HOOKS ==="
# Check TypeScript source count
TS_COUNT=$(ls .claude/hooks/src/*.ts 2>/dev/null | wc -l | xargs)
echo "Found $TS_COUNT TypeScript source files"
# Check bundles exist
BUNDLE_COUNT=$(ls .claude/hooks/dist/*.mjs 2>/dev/null | wc -l | xargs)
echo "Found $BUNDLE_COUNT built bundles"
# Check shell wrappers are executable
FAIL=0
for sh in .claude/hooks/*.sh; do
[ -f "$sh" ] || continue
if [ ! -x "$sh" ]; then
echo "FAIL: Not executable: $sh"
FAIL=$((FAIL+1))
fi
done
SH_COUNT=$(ls .claude/hooks/*.sh 2>/dev/null | wc -l | xargs)
echo "Shell wrappers: $((SH_COUNT - FAIL)) executable, $FAIL need chmod +x"
# Check hooks registered in settings.json exist
echo "Checking registered hooks..."
FAIL=0
# Extract hook commands from settings.json and verify files exist
grep -oE '"command":\s*"[^"]*\.sh"' .claude/settings.json 2>/dev/null | \
sed 's/.*"\([^"]*\.sh\)".*/\1/' | \
sed 's|\$CLAUDE_PROJECT_DIR|.claude|g' | \
sed "s|\$HOME|$HOME|g" | \
sort -u | while read hook; do
# Resolve to actual path
resolved=$(echo "$hook" | sed 's|^\./||')
if [ ! -f "$resolved" ] && [ ! -f "./$resolved" ]; then
echo "WARN: Registered hook not found: $hook"
fi
doneecho "=== MEMORY SYSTEM ==="
# Check DATABASE_URL
if [ -z "$DATABASE_URL" ]; then
echo "FAIL: DATABASE_URL not set"
else
echo "PASS: DATABASE_URL is set"
# Test connection
if psql "$DATABASE_URL" -c "SELECT 1" > /dev/null 2>&1; then
echo "PASS: PostgreSQL reachable"
# Check pgvector
if psql "$DATABASE_URL" -c "SELECT extname FROM pg_extension WHERE extname='vector'" 2>/dev/null | grep -q vector; then
echo "PASS: pgvector extension installed"
else
echo "FAIL: pgvector extension not installed"
fi
# Check table exists
if psql "$DATABASE_URL" -c "\d archival_memory" > /dev/null 2>&1; then
echo "PASS: archival_memory table exists"
# Count learnings
COUNT=$(psql "$DATABASE_URL" -t -c "SELECT COUNT(*) FROM archival_memory" 2>/dev/null | xargs)
echo "INFO: $COUNT learnings stored"
else
echo "FAIL: archival_memory table missing"
fi
else
echo "FAIL: PostgreSQL not reachable"
fi
fi
# Check Python dependencies
echo "Checking Python dependencies..."
(cd opc && uv run python -c "import psycopg2; import pgvector; import sentence_transformers" 2>/dev/null) && \
echo "PASS: Python dependencies available" || \
echo "WARN: Some Python dependencies missing"echo "=== CROSS-REFERENCES ==="
# Check skills reference valid agents
echo "Checking skill → agent references..."
FAIL=0
for skill in $(find .claude/skills -name "SKILL.md"); do
refs=$(grep -oE 'subagent_type[=:]["'\'']*([a-z-]+)' "$skill" 2>/dev/null | sed 's/.*["'\'']//' | sed 's/["'\'']$//')
for ref in $refs; do
if [ -n "$ref" ] && [ ! -f ".claude/agents/$ref.md" ]; then
echo "FAIL: $skill references missing agent: $ref"
FAIL=$((FAIL+1))
fi
done
done
echo "Skill→Agent refs: $FAIL broken"If `--fix` is specified, automatically fix:
1. **Make shell wrappers executable**
chmod +x .claude/hooks/*.sh
2. **Rebuild hooks if TypeScript newer than bundles**
cd .claude/hooks && npm run build
3. **Create
A persistent, learning, multi-agent development environment built on Claude Code Continuous Claude transforms Claude Code into a continuously learning system that maintains context across sessions, orchestrates specialized agents, and eliminates wasting
Repo: parcadei/Continuous-Claude-v3
Automatically trace Claude Code conversations to Braintrust for observability. Captures sessions, conversation turns, and tool calls as hierarchical traces.
Guide for integrating Agentica SDK with Claude Code CLI proxy
Reference guide for Agentica multi-agent infrastructure APIs