docs: document post-processing patterns for eval results (markdown summary, CI integration)

[christso]

Objective

Document how to build post-processing scripts/skills that parse AgentV output artifacts (results.jsonl, results.json, JUnit XML) into presentation formats like markdown summaries, PR comments, and CI job summaries.

Design Latitude

• Add an Astro docs page covering post-processing patterns
• Include example scripts for common use cases:
  • Parse results.jsonl → markdown summary table
  • Parse results.json → GitHub Actions job summary ($GITHUB_STEP_SUMMARY)
  • Parse results.jsonl → ::error:: / ::warning:: annotations
• Optionally create a reusable skill for markdown summary generation
• Keep this outside of AgentV core — presentation is a downstream concern

Acceptance Signals

• Astro docs page with at least 2 example scripts (markdown summary, GitHub job summary)
• Examples show how to read AgentV output formats and produce formatted output
• Skill definition (SKILL.md or similar) for reusable summary generation if applicable

Non-Goals

• No changes to AgentV core output pipeline
• No new built-in output writer
• Not a GitHub Actions reusable workflow (separate scope)

Context

Originally proposed as a built-in Markdown output writer. Reconsidered — markdown summary is a presentation concern better handled downstream. AgentV already generates structured output (JSONL, JSON, JUnit XML). This matches how microsoft/skills does it: their workflow reads results.json and formats the GitHub job summary inline.

See microsoft/skills harness research.

[christso]

Closing — markdown summary is better handled as a downstream skill/script that parses results.jsonl rather than a built-in output writer.

AgentV already generates structured output (JSONL, JSON, JUnit XML). A markdown summary is a presentation concern that belongs in the CI workflow or a post-processing skill, not in core. This matches how microsoft/skills does it — their workflow reads results.json and formats the GitHub job summary inline in the YAML.

A skill or simple script can read results.jsonl and generate any summary format needed (markdown for PR comments, Slack messages, etc.) without coupling it to AgentV's output pipeline.