# Telemetry Report Skill

Track agent activity and report to telemetry system for meta-learning.

## Report Agent Invocation

Note: The actual telemetry collection happens automatically via hooks (SubagentStop hook).
This skill provides utility functions for commands to access telemetry data.

```bash
# Read current session's agent invocations
# The SubagentStop hook automatically records agent names to session state

if [ -n "$SESSION_ID" ]; then
  SESSION_FILE="plugins/psd-claude-coding-system/meta/.session_state_${SESSION_ID}"

  if [ -f "$SESSION_FILE" ]; then
    AGENTS_INVOKED=$(grep "^AGENTS=" "$SESSION_FILE" | cut -d= -f2)
    echo "Agents invoked this session: $AGENTS_INVOKED"
  fi
fi
```

## Query Telemetry for Patterns

```bash
# Check which agents work well together
# Useful for meta-learning and optimization

TELEMETRY_FILE="plugins/psd-claude-coding-system/meta/telemetry.json"

if [ -f "$TELEMETRY_FILE" ] && command -v jq &> /dev/null; then
  # Find most common agent combinations for /work command
  echo "=== Most Common Agent Combinations for /work ==="
  jq -r '.executions[] | select(.command == "work") | .agents_invoked | join(",")' "$TELEMETRY_FILE" \
    | sort | uniq -c | sort -rn | head -5

  # Find average duration by command
  echo -e "\n=== Average Duration by Command ==="
  jq -r '.executions | group_by(.command) | map({command: .[0].command, avg_duration: (map(.duration_ms) | add / length)}) | .[]' "$TELEMETRY_FILE"

  # Find commands with highest success rate
  echo -e "\n=== Success Rates by Command ==="
  jq -r '.executions | group_by(.command) | map({command: .[0].command, success_rate: ((map(select(.success == true)) | length) / length * 100)}) | .[]' "$TELEMETRY_FILE"
fi
```

## Track Parallel Execution

```bash
# When invoking multiple agents in parallel, track the pattern

if [ -n "$SESSION_ID" ]; then
  SESSION_FILE="plugins/psd-claude-coding-system/meta/.session_state_${SESSION_ID}"

  # Mark that this session used parallel execution
  echo "PARALLEL=true" >> "$SESSION_FILE"

  # Track which agents ran in parallel
  echo "PARALLEL_GROUP=$AGENT_LIST" >> "$SESSION_FILE"

  # The Stop hook will read these and add to telemetry.json
fi
```

## Get Recommendations from History

```bash
# Based on current issue/context, get recommendations for which agents to invoke

if [ -f "$TELEMETRY_FILE" ] && command -v jq &> /dev/null; then
  # For similar issues (by keyword), what agents were successful?
  KEYWORDS=$(echo "$ISSUE_TITLE" | tr '[:upper:]' '[:lower:]')

  echo "=== Recommended Agents Based on Similar Issues ==="
  # This is a placeholder - real implementation would use more sophisticated matching
  jq -r ".executions[] | select(.success == true) | select(.command == \"work\") | .agents_invoked[]" "$TELEMETRY_FILE" \
    | sort | uniq -c | sort -rn | head -3
fi
```

## Report Command Metrics

```bash
# At end of command execution, report key metrics for telemetry

echo "=== Command Execution Metrics ==="
echo "Command: $COMMAND_NAME"
echo "Duration: ${DURATION_MS}ms"
echo "Agents Invoked: $AGENTS_INVOKED"
echo "Files Modified: $FILES_MODIFIED"
echo "Tests Run: $TESTS_RUN"
echo "Success: $SUCCESS"

# These metrics are automatically captured by the Stop hook
# which reads from session state and writes to telemetry.json
```

## Usage

### In Commands

```bash
# At start of command
SESSION_ID="${RANDOM}_${RANDOM}"  # Generated by Claude Code
COMMAND_NAME="work"
START_TIME=$(date +%s%3N)

# During execution, agents are invoked
# SubagentStop hook automatically tracks them

# At end of command (Stop hook does this automatically)
END_TIME=$(date +%s%3N)
DURATION_MS=$((END_TIME - START_TIME))

# Stop hook reads session state and updates telemetry.json with:
# - command name
# - duration
# - agents invoked
# - success/failure
# - parallel execution (if applicable)
```

### For Meta-Learning

```bash
# Meta-learning commands can query telemetry for insights
# Include Query Telemetry for Patterns section
# Include Get Recommendations from History section
```