Summary

Adds batched field-value writes to update_project_items using chunked, aliased GraphQL mutations with input-ordered succeeded, failed, or unknown results. One required top-level updated_field applies to every item in the batch.

Why

Implements the MCP portion of github/planning-tracking#3331. It builds on the issue-membership pagination merged in #2914; CLI support will follow separately.

What changed

• Models items[] as three closed oneOf reference variants: node_id, numeric item_id, or item_owner + item_repo + issue_number. The top-level field is a closed ID-or-name variant; null clears it for all items. Tool guidance directs bulk callers here instead of looping update_project_item.
• Resolves the project, shared field, and deduplicated item references before writing. Converts TEXT, NUMBER, DATE, SINGLE_SELECT, and ITERATION values once and rejects duplicate resolved items.
• Sends one mutation kind in sequential chunks of 20 through the existing authenticated githubv4.Client, with a 50-item call cap (at most three requests) and no automatic retries or per-item fallbacks.
• Preserves confirmed partial successes and marks unconfirmed writes unknown when mutation outcomes are ambiguous.

Review guide

1. 101563c0 — internal batch engine and direct parsing, resolution, conversion, and execution tests.
2. 9928ec60 — public projects_write contract/routing, handler and schema coverage, toolsnap, and README.
3. 86810292 — neutral resolution-failure classification with focused fallback and structured-error coverage.
4. a387c39a — five-way bounded concurrent issue-reference resolution with deterministic overlap and cancellation coverage.

Automated tests are 1,603 of 2,602 added lines; production code is 900 and generated output is 99.

MCP impact

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

update_project_items adds reference-only batch items, one shared field/value, node_id references, and structured tri-state results. Existing single-item runtime behavior is unchanged.

Prompts tested (tool changes only)

No live prompts were run. Automated handler tests cover equivalent scenarios:

• Apply one shared update or clear across node, numeric, and paginated issue references, including deduplicated lookups and duplicate-target handling.
• Update 20, 21, and the maximum 50 project items, verifying one, two, and three sequential mutation requests.
• Preserve partial successes and ordered mixed outcomes across GraphQL errors, transport failures, cancellation, and GHES client wiring.

Security / limits

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

The implementation uses the existing injected REST and GraphQL clients and their permissions. The tool is capped at 50 items; numeric and issue-reference reads use at most five concurrent lookup chains, and writes remain sequential in chunks of 20 with no automatic retry.

Tool renaming

• I am renaming tools as part of this PR (e.g. a part of a consolidation effort)
  • I have added the new tool aliases in deprecated_tool_aliases.go
• I am not renaming tools as part of this PR

Note: if you're renaming tools, you must add the tool aliases. For more information on how to do so, please refer to the official docs.

Lint & tests

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

Focused schema, field conversion/resolution, item resolution, mutation execution, and handler tests also pass.

Docs

• Not needed
• Updated (README / docs / examples)