fix(memory): fail fast on stuck external prefetch

[LeonSGP43]

What does this PR do?

Prevents a wedged external memory provider from stalling every turn during prefetch. MemoryManager.prefetch_all() now runs non-builtin provider prefetch calls behind a short fail-fast timeout and skips the provider while the original stuck call is still hung.

Related Issue

Fixes #61800

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)
• ✨ New feature (non-breaking change that adds functionality)
• 🔒 Security fix
• 📝 Documentation update
• ✅ Tests (adding or improving test coverage)
• ♻️ Refactor (no behavior change)
• 🎯 New skill (bundled or hub)

Changes Made

• Add an external-memory prefetch timeout knob to agent/memory_manager.py with a 5s default and HERMES_EXTERNAL_MEMORY_PREFETCH_TIMEOUT override.
• Run external provider prefetch() calls in a daemon thread, returning without blocking when they exceed the timeout.
• Track still-hung prefetch calls per provider and skip repeated turns until the original stuck call returns.
• Add a regression test in tests/agent/test_memory_provider.py that simulates a hung external provider and proves the second turn fails fast instead of re-blocking.

How to Test

1. Run /Users/leongong/Desktop/LeonProjects/worktrees/hermes-agent/.base/.venv/bin/python -m pytest tests/agent/test_memory_provider.py.
2. Run /Users/leongong/Desktop/LeonProjects/worktrees/hermes-agent/.base/.venv/bin/python -m py_compile agent/memory_manager.py.
3. Confirm git diff --check is clean.

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix(scope):, feat(scope):, etc.)
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this fix/feature (no unrelated commits)
• I've run pytest tests/ -q and all tests pass
• I've added tests for my changes (required for bug fixes, strongly encouraged for features)
• I've tested on my platform: macOS 15.5

Documentation & Housekeeping

• I've updated relevant documentation (README, docs/, docstrings) — or N/A
• I've updated cli-config.yaml.example if I added/changed config keys — or N/A
• I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture or workflows — or N/A
• I've considered cross-platform impact (Windows, macOS) per the compatibility guide — or N/A
• I've updated tool descriptions/schemas if I changed tool behavior — or N/A

Screenshots / Logs

• Issue repro shows prefetch() adding a fixed 120s delay per turn when the external HY Memory sidecar hangs.
• This patch changes that failure mode to a one-time timeout per stuck call, then immediate skips until the hanging call returns.

[alt-glitch]

This was generated by AI during triage.

Related: #50800 (competing open PR, 2026-06-22) bounds each external provider's prefetch() with an 8s one-shot-executor timeout — same function (prefetch_all) and same core mechanism (fail-fast timeout on external prefetch). This PR is a superset: it adds per-provider stuck-thread tracking to skip repeated turns while a call is still hung. Fixes #61800. A maintainer should pick the canonical approach across #50800/this one.

[teknium1]

Thanks for addressing the pre-turn stall. The premise is confirmed on current main: agent/turn_context.py:563 calls MemoryManager.prefetch_all() before the tool loop, and agent/memory_manager.py:507 invokes provider prefetch() synchronously.

Problems

• The new HERMES_EXTERNAL_MEMORY_PREFETCH_TIMEOUT override is a non-secret behavioral timeout. AGENTS.md:102-106 requires such settings to be configured through config.yaml, not a new user-facing HERMES_* environment variable.
• The new regression test verifies the repeated-skip state, but after releasing the blocked call it does not issue another prefetch to verify that _stuck_prefetch_threads is cleared and the provider recovers.

Suggested changes

• Preserve the fail-fast implementation but use the existing config path (or retain a fixed default without a user-facing env override).
• Add a post-release recovery assertion to the timeout regression test.

This is an automated hermes-sweeper review.

[teknium1]

Merged via #62290 (commit 8d1c96f) — your stuck-prefetch fail-fast commit landed on main with authorship preserved via rebase-merge (adapted to a bounded internal guard instead of a HERMES_* env var, per our config policy). Fixes #61800. Thanks @LeonSGP43!