> ## Documentation Index > Fetch the complete documentation index at: https://docs.agents-squads.com/llms.txt > Use this file to discover all available pages before exploring further. # Agent Performance > Measure and optimize agent effectiveness with git-based metrics ## Why Measure Performance? Without measurement, you can't answer: * Which agents deliver value? * Where are bottlenecks? * Is quality improving over time? * What's the ROI of agent investment? ## Git-Based Metrics Git provides a natural audit trail for agent work. Every commit, PR, and change is tracked. ### Core Metrics | Metric | Description | Calculation | | ----------------- | -------------------- | ------------------------------------------ | | **Commits/day** | Agent activity level | `git log --author="agent" --since="1 day"` | | **PR merge rate** | Quality of output | Merged PRs / Total PRs | | **Time to merge** | Review efficiency | PR created → PR merged | | **Lines changed** | Scope of work | Additions + Deletions | | **Revert rate** | Error frequency | Reverts / Total commits | ### Tracking Agent Commits Tag agent commits with consistent metadata: ```bash theme={null} git commit -m "feat: implement user auth 🤖 Generated with [Agents Squads](https://agents-squads.com) Co-Authored-By: agents-squads Agent: auth-implementer Squad: engineering Duration: 45m Tokens: 125,000" ``` ### Query Agent Performance ```bash theme={null} # Commits by agent in last week git log --since="1 week" --grep="Agent:" --oneline | wc -l # Agent-specific commits git log --author="agents-squads" --since="1 month" --stat # Commits per agent git log --since="1 month" --grep="Agent:" --format="%s" | \ grep -oP "Agent: \K[^\n]+" | sort | uniq -c | sort -rn # Revert rate echo "scale=2; $(git log --grep="Revert" --since="1 month" | wc -l) / $(git log --since="1 month" | wc -l)" | bc ``` ## Performance Dashboard ### Using `squads` CLI ```bash theme={null} # Overall status squads dashboard # Agent-specific metrics squads feedback stats # View execution history squads memory show engineering ``` ### Custom Metrics Script ```bash theme={null} #!/bin/bash # scripts/agent-metrics.sh SINCE="${1:-1 week}" AGENT_EMAIL="agents@agents-squads.com" echo "=== Agent Performance Report ===" echo "Period: $SINCE" echo "" # Total commits commits=$(git log --author="$AGENT_EMAIL" --since="$SINCE" --oneline | wc -l) echo "Total commits: $commits" # PRs created (requires gh cli) prs_created=$(gh pr list --author="@me" --state=all --json createdAt --jq "length") prs_merged=$(gh pr list --author="@me" --state=merged --json mergedAt --jq "length") echo "PRs created: $prs_created" echo "PRs merged: $prs_merged" echo "Merge rate: $(echo "scale=2; $prs_merged / $prs_created * 100" | bc)%" # Lines changed git log --author="$AGENT_EMAIL" --since="$SINCE" --numstat --pretty="" | \ awk '{add+=$1; del+=$2} END {print "Lines added:", add, "deleted:", del}' # By squad echo "" echo "=== By Squad ===" git log --since="$SINCE" --grep="Squad:" --format="%s" | \ grep -oP "Squad: \K\w+" | sort | uniq -c | sort -rn ``` ## Quality Metrics ### Code Review Scores Track review feedback on agent PRs: ```markdown theme={null} ## PR Review Template ### Agent Output Quality - [ ] Code is correct - [ ] Code follows conventions - [ ] Tests included - [ ] Documentation updated - [ ] No security issues **Score**: 4/5 **Notes**: Minor style issues, otherwise good. ``` ### Automated Quality Checks ```yaml theme={null} # .github/workflows/agent-quality.yml name: Agent Quality Check on: pull_request: branches: [main] jobs: quality: runs-on: ubuntu-latest steps: - uses: actions/checkout@v4 - name: Check if agent PR id: check run: | if git log -1 --format="%b" | grep -q "agents-squads"; then echo "is_agent=true" >> $GITHUB_OUTPUT fi - name: Run quality metrics if: steps.check.outputs.is_agent == 'true' run: | # Lint check npm run lint # Test coverage npm test -- --coverage # Complexity check npx complexity-report src/ - name: Post metrics to PR if: steps.check.outputs.is_agent == 'true' uses: actions/github-script@v7 with: script: | github.rest.issues.createComment({ issue_number: context.issue.number, owner: context.repo.owner, repo: context.repo.repo, body: '## Agent Quality Report\n...' }) ``` ## Benchmarking ### Task Completion Benchmarks Track how long standard tasks take: | Task Type | Target | Current Avg | | ---------------- | ---------- | ----------- | | Bug fix | \< 30 min | 25 min | | Feature (small) | \< 2 hours | 1.5 hours | | Feature (medium) | \< 1 day | 6 hours | | Refactor | \< 4 hours | 3 hours | ### A/B Testing Agents Compare different agent configurations: ```markdown theme={null} ## Experiment: Prompt Optimization **Hypothesis**: Adding domain context improves code quality **Agent A**: Base prompt **Agent B**: Base prompt + CLAUDE.md context **Metrics**: - PR merge rate - Review iterations needed - Time to completion **Results**: - Agent A: 75% merge rate, 2.3 iterations - Agent B: 92% merge rate, 1.1 iterations **Conclusion**: Domain context significantly improves quality ``` ## Monitoring & Alerts ### Performance Thresholds ```yaml theme={null} # .agents/performance.yml thresholds: merge_rate: warning: 0.80 critical: 0.60 revert_rate: warning: 0.05 critical: 0.10 avg_review_iterations: warning: 3 critical: 5 tokens_per_task: warning: 100000 critical: 200000 ``` ### Alerting ```javascript theme={null} // Alert on performance degradation async function checkPerformance() { const metrics = await getAgentMetrics('last_week'); if (metrics.mergeRate < THRESHOLDS.merge_rate.critical) { await alert({ level: 'critical', message: `Agent merge rate dropped to ${metrics.mergeRate}`, squad: metrics.squad }); } if (metrics.revertRate > THRESHOLDS.revert_rate.warning) { await alert({ level: 'warning', message: `High revert rate: ${metrics.revertRate}`, squad: metrics.squad }); } } ``` ## Feedback Loop ### Recording Feedback ```bash theme={null} # After agent task completion squads feedback add engineering # Prompts for: # - Task success (yes/no) # - Quality score (1-5) # - Issues encountered # - Improvement suggestions ``` ### Using Feedback ```bash theme={null} # View feedback trends squads feedback stats # Output: # Squad: engineering # Tasks: 45 # Success rate: 91% # Avg quality: 4.2/5 # Common issues: # - Missing tests (8 occurrences) # - Style inconsistencies (5 occurrences) ``` ### Continuous Improvement ``` Measure → Analyze → Improve → Repeat │ │ │ │ │ └── Update prompts, add context │ └── Identify patterns in failures └── Track metrics over time ``` ## Best Practices * Tag all agent commits with consistent metadata * Track metrics weekly, review monthly * Set quality thresholds and alert on breaches * A/B test prompt and configuration changes * Record human feedback after task completion * Use metrics to guide agent improvements **Measurement pitfalls:** * Optimizing for vanity metrics (commits ≠ value) * Ignoring quality in favor of speed * Not accounting for task difficulty * Missing the feedback loop ## Related Cost-focused metrics Production monitoring