pr-walkthrough

Files (1)

• SKILL.md

SKILL.md (first lines)

1. Record the inspected base and head revisions. Inspect the current PR diff, linked requirements, callers, contracts and existing evidence read-only, within the requested scope. Resolve questions through inspection and identify dependencies. For related PRs, explain their shared goal, then walk each PR separately against its own base. When historical rationale matters, trace relevant commits, PR discussions and linked issues. Cite documented reasons, distinguish them from inference, surface conflicting evidence, and state when the reason remains unknown.
2. Lead with a short end-to-end example and the core decisions; put detailed file and schema accounting afterward. Link the PR and explain the story directly in chat: what happened before, what happens now, and why it matters. Show before/after ASCII flows and distinguish new behavior from extensions, moves and preserved behavior. For complex flows, build the explanation around one concrete scenario; start with its visible effect, then introduce one component at a time in small ASCII diagrams before explaining symbols, storage details and exceptions. Use consistent names and concrete values across prose, examples and diagrams.
3. Walk the important non-test changes in execution order, linking each file to its role. Explain what calls it, where it runs, what it receives/returns or stores, and why its decisions matter. Define unfamiliar terms; use small typed examples, exact storage paths and key code snippets where they clarify the flow. Label process boundaries and data movement in diagrams.
4. Add a separate **Schemas and storage:** section listing every added, modified or removed database schema, collection/table, index, bucket and storage layout. Use a compact table with the change, purpose, relevant fields/types and exact location; distinguish reused resources and mention migrations or retention changes. State `None` when there are no changes.
5. Group plumbing/mechanical files briefly afterward, accounting for every remaining non-test change, including deletions. State relevant configuration/defaults, feature scope and deployment dependencies. Finish with material gaps and a brief distinction between code conclusions and existing runtime evidence.

From AmitRockBear/skills, MIT.

More from AmitRockBear/skills