Agent Persona Exploration - 2026-06-10

[github-actions]

Persona Overview

• Agent: agentic-workflows skill/router
• Scenarios Generated: 10 across 5 personas
• Scenarios Tested: 4 representative scenarios
• Average Quality Score: 4.6/5.0

The available runtime exposed agentic-workflows as a skill/router rather than a promptable custom-agent endpoint. Findings below are based on the loaded skill behavior and referenced .github/aw/*.md workflow guidance.

Key Findings

• Strongest responses are expected for PR review, visual regression, deployment monitoring, and scheduled digest patterns because these are directly covered in .github/aw/*.md references.
• The guidance consistently keeps the main agent job read-only and routes writes through safe-outputs.
• Trigger selection is clear for common cases: pull_request, workflow_run, deployment_status, fuzzy schedule, and slash_command.
• The weakest area is domain-specific report quality for PM/QA use cases, where prompt examples are thinner.

Top Patterns

1. PR tasks usually map to pull_request with scoped paths when possible.
2. Monitoring tasks distinguish in-repo Actions via workflow_run from external services via deployment_status.
3. Common tools are GitHub gh-proxy, playwright for UI checks, and cache-memory for baselines/state.
4. Security practices emphasize read-only permissions, minimal network, safe outputs, and noop for no-op runs.

View High Quality Responses

Scenario	Score	Why it worked well
Backend migration safety	5.0	Dedicated database migration pattern covers trigger, path filters, permissions, safe output, and prompt focus.
DevOps deployment failure	4.6	Clear workflow_run vs deployment_status decision and incident issue pattern.
Frontend visual regression	4.4	Strong playwright + cache-memory guidance with localhost domain restrictions.

View Areas for Improvement

• Product and QA reporting scenarios need more concrete prompt examples for grouping findings by stakeholder impact, owner, and actionability.
• External deployment monitoring could better describe safe ways to include provider log URLs without broad network access.
• The test harness could not prompt a standalone agentic-workflows custom-agent endpoint directly; only the skill/router interface was available.

Recommendations

1. Add persona-specific examples to .github/aw/workflow-patterns.md for QA coverage reports and PM stakeholder digests.
2. Expand .github/aw/deployment-status.md with safe external log-linking guidance and provider-specific context boundaries.
3. Add a compact generated-workflow quality checklist to .github/aw/create-agentic-workflow.md covering trigger fit, tool fit, safe outputs, permissions, network, and prompt clarity.

References: §27290825275

Generated by 🎭 Agent Persona Explorer · 110.2 AIC · ⌖ 2.97 AIC · ⊞ 23.1K · ◷