test(7/N): green up hermes_cli + cron + run_agent + plugins long tail

[Bartok9]

Context

PR #6 (1/N) merged; ~38 pre-existing failures remained on main. This PR cleans up the long tail — 10 failures across 7 files in hermes_cli, cron, run_agent, and plugins. Each was investigated independently; root causes are a mix of genuine production regressions, an intended behavior change, and test-side signature/value drift.

Tests fixed (10)

1. tests/hermes_cli/test_model_validation.py::TestValidateCodexAutoCorrection::test_very_different_name_falls_to_suggestions
2. tests/hermes_cli/test_model_provider_persistence.py::TestProviderPersistsAfterModelSave::test_lmstudio_provider_saved_when_selected
3. tests/hermes_cli/test_model_provider_persistence.py::TestBaseUrlValidation::test_stepfun_provider_saved_with_selected_region
4. tests/hermes_cli/test_tencent_tokenhub_provider.py::TestTencentTokenhubContextLength::test_hy3_preview_context_length
5. tests/cron/test_cron_script.py::TestBuildJobPromptWithScript::test_script_empty_output_noted
6. tests/cron/test_scheduler_mcp_init.py::test_run_job_calls_discover_mcp_tools_before_agent_construction
7. tests/cron/test_scheduler_mcp_init.py::test_run_job_tolerates_discover_mcp_tools_failure
8. tests/run_agent/test_concurrent_interrupt.py::test_running_concurrent_worker_sees_is_interrupted
9. tests/run_agent/test_concurrent_interrupt.py::test_concurrent_interrupt_cancels_pending
10. tests/plugins/test_achievements_plugin.py::test_evaluate_all_stale_cache_serves_stale_and_refreshes_in_background

Root cause per test

#	Test	Root cause	Fix kind
4	hy3-preview context	Round-number-stale assertion. The authoritative source (OpenRouter live metadata, consulted before the hardcoded fallback) reports 262144 = 256*1024 = true binary 256K. Test asserted the decimal 256000.	test value + aligned hardcoded fallback in agent/model_metadata.py
1	very-different codex name	validate_requested_model accepted any codex model name (proxy-friendly catch-all).	prod: reject when no close-match suggestion exists (suggestions empty); keep accept-with-suggestions for near-misses
2	lmstudio persist	Test drift: an existing API key triggers _prompt_api_key's [K]eep/[R]eplace/[C]lear input() prompt, so two prompts fire (K/R/C + base URL), not one. StopIteration on the exhausted side_effect list.	test: provide both input values
3	stepfun region persist	Same root cause as #2 — the K/R/C prompt read stdin in a non-interactive pytest context (OSError: reading from stdin while output is captured).	test: patch builtins.input
5	script empty output	_build_job_prompt returned None on empty script output, so .lower() hit NoneType.	prod: note empty output in the prompt instead of dropping it; the wake-gate / no_agent layers still own the skip-the-AI decision
6, 7	cron MCP init	discover_mcp_tools() was gated behind provider auth, which fails in the credential-stripped test env, so it ran after (or never reached) AIAgent construction.	prod: move MCP discovery before provider resolution (it doesn't depend on the runtime); test: add the resolve_runtime_provider mock the tests were missing (every other run_job test stubs it)
8, 9	concurrent interrupt	Test-fixture drift: the _Stub lacked _tool_guardrails + _append_guardrail_observation (added to _execute_tool_calls_concurrent after the fixture was written), and the fake-tool side_effects didn't accept the messages / pre_tool_block_checked kwargs the production call now passes.	test: update fixture + fake-tool signatures
10	achievements stale-serves-stale	Real race. evaluate_all() returned the _SNAPSHOT_CACHE module global by reference after kicking the background scan; a fast worker rebinds it to fresh data before the return, breaking stale-serves-stale.	prod: capture the snapshot reference before starting the background scan

Validation

pytest -q tests/hermes_cli/ tests/cron/ tests/run_agent/ tests/plugins/:

• This PR's 10 targets: all green.
• 501/501 pass across the directly-touched files (tests/cron/, tests/plugins/test_achievements_plugin.py, tests/hermes_cli/test_tencent_tokenhub_provider.py, test_model_validation.py, test_model_provider_persistence.py, tests/run_agent/test_concurrent_interrupt.py).
• The 19 remaining failures in the four directories belong to the 5 parallel PR clusters (gateway/systemd test_gateway_wsl + test_gateway_service, test_web_server PTY, test_model_switch_custom_providers, test_provider_parity, test_bedrock_model_picker, test_list_picker_providers) plus 2 tests/tools/test_mcp_* import errors — all confirmed pre-existing on main (verified by stashing this PR's diff).

Remaining-on-main count

This PR closes its 10-test cluster from the original 38. The remaining failures are accounted for by the other 5 parallel PRs; together they target all 38.

Ambiguity flags

• feat(egress): host hardening survey — follow-up to #35149 #4 (hy3-preview): judgment call resolved toward 262144. The production comment says "256K context window," 256*1024 == 262144, and the authoritative OpenRouter live metadata returns 262144. The decimal 256000 was the stale round-number form. Updated both the test and the (unreachable-in-practice) hardcoded fallback for consistency.
• feat(egress): CA rotation — follow-up to #35149 #5 (empty script output): changed behavior so empty output is noted rather than silently skipping the AI run. The actual skip-the-AI optimization for empty/no-wake scripts still lives in the wake-gate (_parse_wake_gate) and no_agent paths in run_job; the defensive prompt is None guard in run_job is retained.
• test: green up the Tests suite (1/N) — timezone-independent flow-peak tests #6/test(6/N): update daytona/vercel sandbox cwd assertions to match builtin cd wrapper #7: the MCP-discovery move is a small real improvement (MCP tools now populate even when provider auth uses the fallback chain), not just a test accommodation.

---

Note

Low Risk
Changes are localized (cron prompt/MCP init order, stricter Codex name rejection, context metadata, achievements read path); no auth or payment paths touched.

Overview
This PR fixes a long tail of failing tests (~10) across hermes_cli, cron, run_agent, and plugins, with several production fixes alongside test updates.

Cron: Pre-run scripts that succeed with empty stdout now inject a short “no output” note into the job prompt instead of returning None (which broke callers). discover_mcp_tools() runs before provider resolution so MCP tools register even when auth/fallback would have blocked discovery later.

Models / metadata: hy3-preview context length is aligned to 262144 (binary 256K). OpenAI Codex validation rejects model names with no fuzzy-match suggestions instead of always accepting unknown slugs.

Achievements plugin: On stale cache + background refresh, the API snapshots the in-memory cache before starting the scan so concurrent updates don’t return fresh data when stale was intended.

Tests: Provider persistence flows stub extra input() prompts; cron MCP tests mock resolve_runtime_provider; concurrent-interrupt stubs add tool guardrails and updated fake-tool signatures.

^{Reviewed by Cursor Bugbot for commit add2e05. Bugbot is set up for automated code reviews on this repo. Configure here.}

[github-actions]

🔎 Lint report: fix/test-suite-green-7n-cli-cron-misc 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: 7749 on HEAD, 7746 on base (🆕 +3)

🆕 New issues (6):

Rule	Count
invalid-argument-type	5
unresolved-attribute	1

First entries

tests/run_agent/test_concurrent_interrupt.py:94: [unresolved-attribute] unresolved-attribute: Unresolved attribute `_append_guardrail_observation` on type `_Stub`
run_agent.py:12539: [invalid-argument-type] invalid-argument-type: Argument to function `_is_oauth_token` is incorrect: Expected `str`, found `str | dict[Unknown, Unknown] | Any | ... omitted 3 union elements`
run_agent.py:6649: [invalid-argument-type] invalid-argument-type: Argument to function `build_anthropic_client` is incorrect: Expected `str`, found `str | dict[Unknown, Unknown] | Any | ... omitted 3 union elements`
tests/run_agent/test_concurrent_interrupt.py:41: [invalid-argument-type] invalid-argument-type: Argument to constructor `NonCallableMock.__new__` is incorrect: Expected `bool | None`, found `MagicMock`
tests/run_agent/test_concurrent_interrupt.py:41: [invalid-argument-type] invalid-argument-type: Argument to constructor `NonCallableMock.__new__` is incorrect: Expected `bool`, found `MagicMock`
run_agent.py:12542: [invalid-argument-type] invalid-argument-type: Argument to function `len` is incorrect: Expected `Sized`, found `(str & ~AlwaysFalsy) | (dict[Unknown, Unknown] & ~AlwaysFalsy) | (Any & ~AlwaysFalsy) | ... omitted 3 union elements`

✅ Fixed issues (4):

Rule	Count
invalid-argument-type	3
invalid-return-type	1

First entries

cron/scheduler.py:814: [invalid-return-type] invalid-return-type: Return type does not match returned value: expected `str`, found `None`
run_agent.py:6649: [invalid-argument-type] invalid-argument-type: Argument to function `build_anthropic_client` is incorrect: Expected `str`, found `str | dict[Unknown | str, Unknown | str | dict[str, str]] | Any | ... omitted 3 union elements`
run_agent.py:12539: [invalid-argument-type] invalid-argument-type: Argument to function `_is_oauth_token` is incorrect: Expected `str`, found `str | dict[Unknown | str, Unknown | str | dict[str, str]] | Any | ... omitted 3 union elements`
run_agent.py:12542: [invalid-argument-type] invalid-argument-type: Argument to function `len` is incorrect: Expected `Sized`, found `(str & ~AlwaysFalsy) | (dict[Unknown | str, Unknown | str | dict[str, str]] & ~AlwaysFalsy) | (Any & ~AlwaysFalsy) | ... omitted 3 union elements`

Unchanged: 4072 pre-existing issues carried over.

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

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is ON, but it could not run because the branch was deleted or merged before autofix could start.}

^{Reviewed by Cursor Bugbot for commit add2e05. Configure here.}

[cursor]

Redundant "no output" repeated in prompt message

Low Severity

The constructed prompt reads "The pre-run script ran successfully but produced no output (no output)." — the parenthetical (no output) is a verbatim repetition of what the sentence already states. This looks like an editing artifact where the parenthetical was likely meant to contain a different clarification (e.g., "stdout was empty" or "exit code 0") but accidentally duplicated the same phrase.

 

^{Reviewed by Cursor Bugbot for commit add2e05. Configure here.}