Handle persona-explorer invocation failures as first-class partial results

[Copilot]

The persona explorer run surfaced a tooling failure in threat detection and produced useful-but-partial results without scenario scoring. This updates the workflow guidance so evaluation-mode invocation failures are treated as expected partial outcomes, with explicit N/A scoring semantics.

• Ad hoc evaluation invocation guidance

  • Updated .github/aw/github-agentic-workflows.md to add explicit research/evaluation invocation phrasing for no-file-creation mode.
  • Clarifies how to ask for recommendation-only output from agentic-workflows.

• Fallback evaluation behavior when custom-agent invocation is unavailable

  • Updated .github/aw/create-agentic-workflow.md to define fallback behavior:
    • continue using local gh-aw guidance,
    • report partial results,
    • mark scoring/response quality unavailable.

• Persona explorer partial-results scoring contract

  • Updated .github/workflows/agent-persona-explorer.md to standardize failure handling:
    • mark failed scenarios as invocation_unavailable,
    • set quality scoring to N/A,
    • exclude unavailable scenarios from numeric averages,
    • allow Average Quality Score to be N/A when scoring is unavailable,
    • explicitly frame invocation failure as a normal partial-results outcome in final issue output.

- If invocation fails: mark the scenario as `invocation_unavailable`, set quality scoring to `N/A`, and continue.
- If scenario invocation failed, note that scoring is unavailable for affected scenarios and exclude them from the numeric average.
- **Average Quality Score**: [X.X/5.0 or N/A when invocation/scoring is unavailable]

• Fixes Agent Persona Exploration - 2026-07-31 #49373

[Copilot]

Pull request overview

Updates persona evaluation guidance to treat unavailable custom-agent invocations as partial results with unavailable scoring.

Changes:

• Adds recommendation-only evaluation phrasing and fallback guidance.
• Defines invocation_unavailable and N/A scoring behavior.
• Updates the compiled workflow metadata.

Show a summary per file

File	Description
.github/aw/create-agentic-workflow.md	Adds fallback evaluation guidance.
.github/aw/github-agentic-workflows.md	Adds explicit research invocation phrasing.
.github/workflows/agent-persona-explorer.md	Defines partial-result and scoring behavior.
.github/workflows/agent-persona-explorer.lock.yml	Updates the generated body hash.

Review details

Tip

Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Suppressed comments (1)

.github/workflows/agent-persona-explorer.md:225

• This leaves mixed-success reporting ambiguous: line 179 requires averaging the successfully scored scenarios, while this placeholder permits N/A whenever any invocation is unavailable. The preceding field also conflates the 3–4 selected scenarios with scenarios that actually returned a response, which would misreport an all-unavailable run as tested. Separate the counts and use N/A only when no scenarios were scored.

- **Average Quality Score**: [X.X/5.0 or N/A when invocation/scoring is unavailable]

• Files reviewed: 4/4 changed files
• Comments generated: 1
• Review effort level: Balanced

[github-actions]

🎉 This pull request is included in a new release.

Release: v0.84.2