Add response field filtering to pull_request_read reviews and check runs - #3310

Summary

Adds optional fields selection to pull_request_read for the get_reviews and get_check_runs methods, allowing callers to request only the response fields they need.

Omitting fields, or passing an empty array, preserves the existing full response.

Why

pull_request_read currently returns the full payload for every method, even when callers only need a small subset of the data.

This is particularly expensive for:

• get_reviews, where review bodies can contain several KB of markdown even when the caller only needs review state and reviewer information.
• get_check_runs, where callers often only need fields such as check name and conclusion.

This addresses the high-volume cases described in #3286 while keeping response filtering opt-in and backward-compatible with the existing fields behavior elsewhere in the server.

Unlike the issue's suggested default omission of review bodies, this PR does not change the default response shape: existing callers continue to receive the full response unless they explicitly request a subset of fields.

Refs #3286

get_files is intentionally not included because #3242 already implements field selection for that method.

What changed

• Added an optional fields parameter to the consolidated pull_request_read schema.

• Added selectable fields for:

  • get_reviews
  • get_check_runs

• Added method-aware validation so fields valid for one method cannot be used with another.

• Validation happens before any GitHub API request is made.

• Reused the existing response-field filtering helpers and fields telemetry.

• Preserved total_count when filtering individual get_check_runs entries.

• Kept lockdown filtering ahead of response projection for pull request reviews.

• Preserved existing IFC labeling behavior.

• Added regression coverage for schema, filtering, validation, telemetry, lockdown composition, and IFC labels.

• Updated the generated tool snapshot and README documentation.

Because pull_request_read is a consolidated tool, its JSON schema exposes the union of selectable fields for the supported methods. Runtime validation then narrows that set according to the selected method.

The per-method field enums are also used to derive the schema union, keeping the advertised schema and runtime validator from drifting apart.

MCP impact

• No tool or API changes
• Tool schema or behavior changed
• New tool added

Adds the optional fields parameter for pull_request_read when using get_reviews or get_check_runs.

Existing calls without fields, including calls with an empty fields array, retain the previous full response.

Example:

{
  "method": "get_reviews",
  "owner": "owner",
  "repo": "repo",
  "pullNumber": 42,
  "fields": ["state", "user"]
}

returns only the selected fields for each review.

Similarly:

{
  "method": "get_check_runs",
  "owner": "owner",
  "repo": "repo",
  "pullNumber": 42,
  "fields": ["name", "conclusion"]
}

keeps the response envelope, including total_count, while filtering each check run.

Prompts tested (tool changes only)

Representative workflows covered by the implementation and regression tests:

• “Show the review states and reviewers for this pull request without review bodies.”

  • Uses get_reviews with fields: ["state", "user"].

• “Show only the check names and conclusions for this pull request.”

  • Uses get_check_runs with fields: ["name", "conclusion"].

• Existing calls without fields.

  • Continue returning the complete review/check-run payload.

• Invalid cross-method selections.

  • For example, get_reviews with fields: ["name"] is rejected before making a GitHub API request.

Security / limits

• No security or limits impact
• Auth / permissions considered
• Data exposure, filtering, or token/size limits considered

This change only removes unrequested fields from responses; it does not expose any data that was previously unavailable.

For get_reviews, lockdown filtering continues to run before response field filtering. Tests also verify that IFC metadata remains unchanged when projection is enabled.

The existing low-cardinality fields telemetry is reused with tool=pull_request_read; no new telemetry dimensions are introduced.

Tool renaming

• I am renaming tools as part of this PR

  • I have added the new tool aliases in deprecated_tool_aliases.go

• I am not renaming tools as part of this PR

Lint & tests

• Linted locally with ./script/lint
• Tested locally with ./script/test

Additional verification:

go test -race -count=1 ./...  → exit 0
./script/lint                  → exit 0, 0 issues
git diff --check              → clean

Docs

• Not needed
• Updated (README / generated tool schema snapshot)

The pull_request_read tool documentation now describes the optional fields parameter and its supported methods.