fix(config): surface invalid platform_toolsets instead of silently dropping tools (#38798)

[kshitijk4poor]

Summary

A corrupted or hand-edited platform_toolsets entry no longer silently disables all tools — invalid toolset names are now surfaced loudly.

Root cause (#38798)

When platform_toolsets contains an invalid toolset name (the reported case: a migration rewrote hermes-cli → the non-existent hermes), resolve_toolset() returns [] for that name. The platform ends up with zero tools, the system prompt is built without tool guidance, and the agent silently degrades to text-only replies — no error, no warning, no log entry. The cause took significant debugging to find.

Fix

Surface the failure at two points (hermes_cli/):

• After migration (migrate_config) — validate platform_toolsets and record/print a warning per unknown name, with a hermes-<platform> suggestion when that would have been valid (the exact bug: config migration (v25->v26) corrupts platform_toolsets, silently kills all tools #38798 shape: cli: [hermes] → "did you mean 'hermes-cli'?").
• At runtime (_get_platform_tools) — if a platform was explicitly configured but every toolset name is invalid, log a warning when tools are resolved for a session. This catches an already-corrupted config at startup, not only on the next hermes update.

The logic lives in a new pure, side-effect-free helper (hermes_cli/toolset_validation.py) with validate_toolset injected, so it's unit-testable without the tool registry or launching Hermes.

Scope note (verify-before-build)

The original v25→v26 migration that caused the corruption no longer exists — config format is now v30, and a full-history pickaxe shows no migration step that rewrites toolset names. So the prevent PR (#39581) is obsolete and the repair-migration PRs (#41006/#40208) target a corruption main no longer produces. This change is the durable defense against the silent-failure mode regardless of cause, which is exactly the issue's stated "Expected: log a warning."

Validation

• tests/hermes_cli/test_toolset_validation.py (new) + test_tools_config.py — 113 passed, no regression.
• Coverage: the bug: config migration (v25->v26) corrupts platform_toolsets, silently kills all tools #38798 corruption shape (warns + suggests hermes-cli), mixed valid/invalid, zero-tools end state, non-dict / scalar / non-string entries, a real-validate_toolset integration test, and a runtime caplog test.
• Mutation-verified load-bearing: neutering the helper warning fails the bug: config migration (v25->v26) corrupts platform_toolsets, silently kills all tools #38798 test; neutering the runtime guard fails the runtime caplog test.
• ruff clean.

Salvaged from #39207 by @lEWFkRAD (authorship preserved via cherry-pick; AUTHOR_MAP entry added).

Closes #38798. Supersedes #39581 (prevent-in-v25→v26 — that path is gone), #41006 / #40208 (repair-migration for already-corrupted configs).

[github-actions]

🔎 Lint report: salvage/38798-toolset-validation vs origin/main

ruff

Total: 0 on HEAD, 0 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 0 pre-existing issues carried over.

ty (type checker)

Total: 11484 on HEAD, 11485 on base (✅ -1)

🆕 New issues (2):

Rule	Count
invalid-assignment	1
unresolved-import	1

First entries

tests/run_agent/test_credits_notices_toggle.py:76: [invalid-assignment] invalid-assignment: Object of type `None` is not assignable to attribute `_credits_session_start_micros` of type `int`
tests/hermes_cli/test_toolset_validation.py:7: [unresolved-import] unresolved-import: Cannot resolve imported module `pytest`

✅ Fixed issues (2):

Rule	Count
unresolved-attribute	2

First entries

run_agent.py:3002: [unresolved-attribute] unresolved-attribute: Object of type `Self@get_credits_spent_micros` has no attribute `_credits_session_start_micros`
tests/run_agent/test_credits_notices_toggle.py:76: [unresolved-attribute] unresolved-attribute: Unresolved attribute `_credits_session_start_micros` on type `AIAgent`

Unchanged: 6032 pre-existing issues carried over.

Diagnostics are surfaced as warnings — this check never fails the build.