Feat: warn when a skill's resource payload exceeds the recommended size limit (adk-python parity)

[AmaadMartin]

Please ensure you have read the contribution guide before creating a pull request.

Link to Issue or Description of Change

1. Link to an existing issue (if applicable):

No existing issue.

2. Or, if no issue exists, describe the change:

Problem: getSkillResourceFiles() ships a skill's whole references/, assets/ and scripts/ payload into a code executor on every script call. The TypeScript SDK measures none of it, so an oversized skill fails only later, inside the executor, with an opaque message. adk-python accounts for the same payload and warns above a 16 MB soft limit.

Solution: Sum the byte length of every resource that is actually shipped, and log one warning when the total exceeds MAX_SKILL_PAYLOAD_BYTES. The limit stays soft, as in Python: nothing is rejected, truncated or thrown, and the returned File[] is unchanged. The accumulator runs after the existing fileContent === undefined guard, so a skipped resource contributes 0 bytes.

Deliberate deviations from the reference:

1. The count uses Buffer.byteLength, not String.prototype.length. Python's len() over a str counts code points, so the reference undercounts multi-byte text by up to 3x. Parity here is the behaviour, and a byte count is the correct implementation of it. No adk-python change is made.
2. Python names the constant _MAX_SKILL_PAYLOAD_BYTES. This repo forbids the leading underscore, so it is MAX_SKILL_PAYLOAD_BYTES. The value and the message wording are identical.
3. Buffer.byteLength accepts a string and a Buffer and returns the byte length of each, so one call replaces a Buffer.isBuffer ternary.

The constant is not added to index.ts or common.ts. This is not new public API.

Testing Plan

Unit Tests:

• I have added or updated unit tests for my change.
• All unit tests pass locally.

Six cases in a new describe('payload size accounting', ...) in core/test/tools/skills/run_skill_script_tool_test.ts. No existing test is modified, removed or weakened.

npx vitest run --project unit:core core/test/tools/skills/run_skill_script_tool_test.ts
# Test Files 1 passed (1) / Tests 14 passed (14)

Coverage of the new code is 100% of lines and branches, read from coverage-final.json. The file's remaining uncovered lines (200-201, 216-217) predate this change. No coverage suppression was added.

Mutation proofs. I mutated one source line per row, re-ran the targeted suite, then restored the source:

Mutation	Test that fails	Failure message
Buffer.byteLength(fileContent) -> fileContent.length	measures multi-byte text in UTF-8 bytes, not UTF-16 code units	expected "warn" to be called 1 times, but got 0 times
totalBytes > -> totalBytes >=	does not warn when skill resources are exactly at the limit	expected "warn" to not be called at all, but actually been called 1 times
delete the logger.warn call	the four warning cases	expected "warn" to be called 1 times, but got 0 times
totalBytes += -> totalBytes =	sums bytes across resource types	expected "warn" to be called 1 times, but got 0 times

The first mutation is why the multi-byte case exists. Every single-byte test passes against a String.length count, so those tests give no signal on the defect this feature guards against. The fixture is 8,388,609 'é' characters: under the limit by String.length, over it by UTF-8 byte length.

This change adds no error path. getSkillResourceFiles does not throw and has no errorCode. The negative path is "no warning", pinned by the under-limit and at-limit cases.

Manual End-to-End (E2E) Tests:

No E2E test is included. The change alters neither the executor payload nor any tool response, so there is no cross-component contract to exercise. To check the log by hand, set the ADK log level to warn and run a skill script whose resources exceed 16 MiB. The warning renders once:

WARN: [ADK] Skill 'research-helper' resources total 17825834 bytes, exceeding the recommended limit of 16777216 bytes.

Repo gates on the pushed commit: npm run build passes, npx eslint on both touched files is clean, and npx prettier --check reports All matched files use Prettier code style!. npm run ts:check is already red on main (280 errors). This diff adds zero new errors: the count is 280 before and after, and the one error in the touched test file is the pre-existing as File cast in a test I did not change.

Collision check. I listed all 697 fork PRs. Ten open PRs touch run_skill_script_tool.ts (#634, #564, #556, #518, #517, #481, #410, #356, #353, #298), but none adds payload accounting: grep -icE 'byteLength|PAYLOAD_BYTES|payload size' over each diff returns 0. The overlap is textual only, in runAsync, so this branches from main rather than stacking.

Checklist

• I have read the CONTRIBUTING.md document.
• I have performed a self-review of my own code.
• I have commented my code, particularly in hard-to-understand areas.
• I have added tests that prove my fix is effective or that my feature works.
• New and existing unit tests pass locally with my changes.