Fixes #415
Refs #513

What does this PR do?

This PR adds insertion-time tool result compaction so large tool outputs are reduced before they are appended back into the model conversation. The goal is to keep recent tool calls useful without letting raw HTML, long terminal output, or large JSON/file payloads dominate the next prompt before normal context compression has a chance to run.

The approach is deterministic and cache-friendly: it compacts only the tool-result message being appended, leaves TUI/log callbacks on the raw tool result, and avoids LLM summaries or external vector storage.

Related Issue

Fixes #415
Refs #513

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 agent/tool_context_policy.py with deterministic compaction for terminal, file, web/HTML, JSON, and generic tool results.
• Route both sequential and concurrent tool execution through the same context-only compaction path before make_tool_result_message().
• Keep display/log callbacks on the raw tool result so user-visible output is not reduced by this policy.
• Add tool_context defaults to hermes_cli/config.py and cli-config.yaml.example.
• Document tool context compaction in website/docs/user-guide/configuration.md.
• Add focused tests in tests/agent/test_tool_context_policy.py.

How to Test

1. Run the focused regression suite:

   /Users/zhi/.hermes/hermes-agent/venv/bin/python -m pytest -q \
     tests/agent/test_tool_context_policy.py \
     tests/tools/test_tool_result_storage.py \
     tests/tools/test_budget_config.py \
     tests/agent/test_context_compressor.py

2. Compile the touched modules:

   /Users/zhi/.hermes/hermes-agent/venv/bin/python -m py_compile \
     agent/tool_context_policy.py \
     agent/tool_executor.py \
     hermes_cli/config.py

3. Local smoke test on macOS with a Chinese-language Hermes session and local llama.cpp backend: a real 57,701-character HTML/tool result compacted to an 823-character model-context result while preserving title, description, headings, and text preview.

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, local llama.cpp backend, Chinese-language Hermes session

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

For New Skills

N/A

Screenshots / Logs

Focused tests:

158 passed in 17.16s