[plan] Document pre-step data-fetching pattern in create-agentic-workflow.md

[github-actions]

Objective

Document the "pre-step data-fetching" pattern in .github/aw/create-agentic-workflow.md as a first-class best practice with a reusable template.

Context

From discussion #24656 (Agent Persona Exploration 2026-04-05): 3 of 8 tested scenarios independently recommended downloading heavy data (logs, CI artifacts, build output) before the AI session begins. This saves tokens and improves reliability. The pattern should be explicitly documented so users adopt it proactively rather than discovering it later.

Approach

1. Open .github/aw/create-agentic-workflow.md
2. Add a new section "Pre-step Data Fetching" (or integrate into an existing "Performance" or "Best Practices" section)
3. The section should:
   • Explain why: heavy data in the AI prompt wastes tokens and slows responses
   • Show a concrete example using a run: step before the agent step to download/prepare data (e.g., downloading workflow logs, fetching artifacts, running npm ci && npm run build)
   • Provide a reusable template snippet
   • Give 2–3 example use cases (deployment logs, build artifacts, test results)
4. Cross-reference the repo-memory docs for the related "load baseline before AI" pattern

Files to Modify

• .github/aw/create-agentic-workflow.md — add pre-step data-fetching section/pattern

Acceptance Criteria

• A "pre-step data fetching" pattern section exists in create-agentic-workflow.md
• The explanation covers why this saves tokens
• A concrete YAML template snippet is provided
• At least 2 example use cases are described (e.g., logs, artifacts)
• Cross-reference to repo-memory for baseline loading is included

Generated by Plan Command for issue #discussion #24656 · ● 321.2K · ◷

• expires on Apr 7, 2026, 8:30 AM UTC