[FIX] PR #1550 CI failures — _is_unattended_context too aggressive in test env

[Lexus2016]

Problem

PR #1550 (suppress pending_approval retries in cron/subagent contexts) has CI failures on test slices 2/8 and 8/8.

Root cause: _is_unattended_context() returns True whenever there's no interactive CLI, no gateway, and no cron session. In the CI test environment (non-interactive, non-gateway, non-cron), this evaluates to True, causing check_all_command_guards and check_execute_code_guard to return blocked instead of pending_approval. Existing tests that expect pending_approval behavior break.

Proposed Fix

Make _is_unattended_context() require a positive signal of an unattended context (HERMES_CRON_SESSION or HERMES_SUBAGENT), instead of the catch-all "nobody's home" default. This preserves the fix for actual cron/subagent sessions while not affecting tests or interactive sessions that haven't set HERMES_INTERACTIVE.

def _is_unattended_context() -> bool:
    """True when in a cron or subagent context where no human can approve."""
    if env_var_enabled("HERMES_CRON_SESSION"):
        return True
    if env_var_enabled("HERMES_SUBAGENT"):
        return True
    return False

Scope

• ~5-line change to _is_unattended_context() in tools/approval.py
• Update test to match new behavior
• Fits ≤200-line cap

[Lexus2016]

Implemented in PR #1556. The fix requires a positive cron/subagent signal (contextvar) rather than the catch-all default that broke CI. Over the 200-line autonomous-merge cap due to the tightly-coupled contextvar mechanism across 3 files — holding for human review.