fix(mcp): capText's truncation marker could push output past budget

[teyrebaz33]

Primary change type

• Bug fix
• Documentation
• Feature
• Tests
• Dependency update
• Maintenance or refactor

Problem and motivation

execution-projection.ts documents capText (packages/mcp/src/tools/shared.ts) as a "hard char budget" primitive: "the result is bounded for ANY argument combination." The implementation didn't honor that — it sliced text to exactly budget characters, then appended a truncation marker AFTER the slice, so the returned string was budget + marker.length characters, not budget. Concretely, capText(text, 2000, someUrl) could return a 2094-char string.

Summary and scope

Since the marker's own length depends on the dropped-char count, which depends on where we slice, which depends on the marker's length, this resolves it with a small converging loop that shrinks the slice point until slice + marker fits within budget, with a final .slice(0, budget) backstop for degenerate tiny budgets. This also fixes the reported dropped-char count, which the old version could get slightly wrong for the same reason.

Out of scope: no changes to execution-projection.ts itself or to the field budgets it configures; no dependency changes.

Related work

Related issue or discussion: #653

Validation

# packages/mcp: vitest run / tsc --noEmit / eslint src --ext .ts
15/15 test files, 107/107 tests passing (including the new suite)
tsc --noEmit: clean
eslint: 0 errors, 0 warnings

Tests and documentation

Added packages/mcp/src/tools/shared.test.ts (no test file existed for this module before): a 240-case brute-force sweep across budgets (including 0/1/5/10, and the real PREVIEW_BUDGET/DEFAULT_FIELD_BUDGET values) confirming the fix never exceeds budget, plus a regression test verified via git stash to fail 4/6 cases against the pre-fix implementation and pass 6/6 against the fix. No user-facing documentation changes needed.

Compatibility and release impact

• Breaking or externally visible changes: None. capText's signature and return type are unchanged; only truncated outputs near a field's budget boundary get slightly shorter (by the marker's length) to actually respect the documented bound.
• Changeset: Added (.changeset/fix-captext-budget-overrun.md), patch bump.

Security

• I have not included secrets, credentials, private data, or unsanitized logs.
• This pull request does not publicly disclose a suspected vulnerability. I
  will follow the
  Security Policy for
  private reporting.

AI assistance

• I did not use AI assistance for this change.
• I used AI assistance and have described it below.

I used Claude (Anthropic) as a coding assistant throughout: it helped find the bug (including writing a small script that proved the overrun with real numbers), draft the fix and the brute-force test sweep, and run the verification commands quoted above. I reviewed and ran every command myself, read and understood the resulting diff line by line, and can explain and maintain every change in this PR.

Checklist

• I read CONTRIBUTING.md, and this contribution follows the direct-PR or issue-first policy.
• This pull request addresses one focused problem and contains no unrelated cleanup.
• I added or updated tests, or explained above why tests are not applicable.
• I ran the relevant build, typecheck, lint, and test commands, or explained
  any N/A checks above.
• I updated documentation for user-facing changes, or marked it N/A above.
• I added a Changeset for a published-package change, or explained why it is not applicable.
• I can explain and maintain every submitted change, including any AI-assisted work.