Agent Persona Exploration - 2026-04-25

[github-actions[bot]]

Persona Overview

• Methodology: Pattern analysis of 201 existing workflows + create-agentic-workflow.md guide (GitHub Copilot Chat agent not directly invocable from workflow context)
• Scenarios Tested: 4 representative scenarios (one per key persona)
• Average Quality Score: 4.2/5.0
• Run: §24921767031

---

Personas & Scenarios

Persona	Scenario	Trigger	Engine	Score
Backend Engineer	DB schema migration review on PRs	pull_request	copilot	4.2/5
DevOps Engineer	Monitor failed deployments, create RCA incidents	schedule every 6h	claude	4.8/5
QA Tester	Test coverage analysis on PRs, comment recommendations	pull_request	copilot	3.8/5
Product Manager	Weekly digest of completed features by customer impact	schedule weekly	copilot	4.6/5

---

Key Findings

• Schedule triggers dominate (139/201 workflows use schedule), suggesting the agent excels at periodic automation
• 85% of workflows use safe-outputs correctly — the security model is well-internalized
• copilot engine preferred (102/201) but claude is favored for complex investigation tasks (aw-failure-investigator, api-consumption-report)
• Issue-triggered workflows outnumber PR-triggered (177 vs 85) — PR-based code quality workflows are underrepresented in templates
• [default] toolset is the most common (43 uses) — agent tends toward breadth over minimal permissions

---

Top Patterns

1. Triggers: Schedule-daily/weekly > issue-opened > pull_request-opened > workflow_dispatch fallback
2. Tools: github: toolsets: [default] + bash: ["*"] for power workflows; scoped toolsets (repos, issues) for simpler ones
3. Security: safe-outputs used consistently; sandbox: agent: awf adopted in ~17 workflows for network firewall enforcement
4. Outputs: create-issue (61 uses), create-pull-request (40), add-comment (39), create-discussion (21)

View High Quality Responses (Top 2)

DevOps - Deployment Failure Monitoring (4.8/5)

• Near-perfect match: aw-failure-investigator workflow uses every 6h schedule, claude engine, agentic-workflows MCP tool, create-issue with group: true and labels — ideal for incident tracking with deduplication.
• Security: bash: ["*"] granted because log analysis requires broad access; compensated with actions: read only permissions.

PM - Weekly Feature Digest (4.6/5)

• Excellent match: weekly-issue-summary and api-consumption-report show strong scheduled digest patterns with chart generation.
• Uses edit: tool + bash for data processing, create-discussion for output (correct — digests are informational, not task-tracking).

View Areas for Improvement (Top 2)

QA Test Coverage Analysis (3.8/5)

• No direct template for PR-triggered test coverage analysis exists in the workflow library.
• Agent would likely adapt breaking-change-checker pattern (daily schedule → PR trigger), but guidance on how to extract coverage data from CI artifacts is missing.
• The actions: read permission + toolsets: [actions] combination needed for artifact access isn't well-documented for PR workflows.

Tool Scope Inconsistency

• Some workflows use toolsets: [all] (3 instances) when narrower scopes would suffice — signals the agent may over-provision tool access when uncertain about requirements.
• Recommend stronger guidance in create-agentic-workflow.md on minimal toolset selection.

---

Recommendations

1. Add a test coverage analysis shared import in .github/aw/imports/ — a coverage-analysis-base.md template covering how to access CI artifacts and comment on PRs would fill the QA persona gap.

2. Strengthen toolset selection guidance in .github/aw/create-agentic-workflow.md — add a decision table: when to use [default] vs [repos] vs [issues] vs [all], with examples tying toolset choice to task type.

3. Document engine selection heuristics in .github/aw/github-agentic-workflows.md — clarify when to prefer claude over copilot (e.g., multi-step investigation requiring long context and reasoning → claude; fast PR comments → copilot).

---

References: §24921767031

Generated by Agent Persona Explorer · ● 782.3K · ◷

[github-actions[bot]]

This discussion has been marked as outdated by Agent Persona Explorer.

A newer discussion is available at Discussion #28538.