Problem

The pr-report skill generates weekly PR reports for HyperShift repositories, but the current pipeline has gaps that limit its usefulness:

Limited analysis depth: Reports rely solely on PR titles and descriptions. There is no mechanism to fetch actual code diffs and analyze what changed, making it difficult to assess real impact, identify breaking changes, or verify alignment between descriptions and code.

No prioritization for analysis: When dozens of PRs merge in a week, there is no way to identify which ones are most important to analyze in depth. All PRs are treated equally regardless of Jira priority, type of change, or strategic alignment.

Sub-agent sandboxing limitations: The deep analysis pipeline relies on Task agents reading and writing files in locations they cannot access, causing failures during automated runs.

Inaccurate contributor attribution: When analyzing PRs at scale, author information is sometimes inferred from PR content rather than sourced from structured data, leading to misattribution.

No narrative output format: The existing reports (data metrics + impact analysis) serve operational needs but lack a reader-friendly narrative format suitable for sharing with broader audiences as a progress update or blog post.

Single-repo coverage: The script only covered openshift/hypershift and openshift-eng/ai-helpers, missing CI and test configuration changes in openshift/release that are relevant to the project.

Objectives

_ _Deep code analysis mode*: Add a --deep flag that fetches actual code diffs for selected PRs and analyzes them with Task agents to identify breaking changes, API modifications, test coverage, and alignment between descriptions and reality.
_ _PR scoring and selection*: Provide a scoring system that ranks PRs by importance (Jira priority, change type, strategic alignment) so users can focus deep analysis on the most impactful work.
_ _Reliable agent pipeline*: Ensure the end-to-end deep analysis flow (fetch diffs → analyze with agents → aggregate results) works without manual intervention by addressing file access and write constraints in the agent execution model.
_ _Accurate attribution*: Guarantee that contributor credit is always sourced from authoritative PR metadata rather than heuristics.
_ _Narrative progress report*: Provide an opt-in --progress-report format that tells the story of the week's work with technical depth, thematic grouping, and problem-first storytelling — modeled after the Dolphin Emulator progress blog format.
_ _Cross-repo coverage*: Extend PR fetching to include openshift/release with path-based filtering for HyperShift-related changes.