A diagnostic and patching skill designed to repair Hermes Agent runs that silently stall due to premature control-loop stops.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install hermes-continuation-maintainer
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install hermes-continuation-maintainer using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Hermes Continuation Maintainer?
The Hermes Continuation Maintainer is a technical utility designed to keep Hermes Agent runs flowing smoothly without requiring manual intervention. When utilizing Openclaw Skills, developers often encounter a specific bottleneck where the provider returns a "finish_reason=stop" for a non-final status message. This causes the conversation loop to incorrectly treat the turn as terminal, leaving the agent silently waiting despite having active work left to do.
This skill addresses this problem directly by analyzing agent logs and runtime databases, classifying stalls, and applying a targeted patching policy to the agent's core helper and conversation files. By implementing precise continuation heuristics, it ensures the agent proceeds with necessary tool calls instead of stopping prematurely, significantly increasing autonomous run efficiency.
Hermes Continuation Maintainer Use Cases
- The Hermes Agent silently stops after outputting a brief, non-final status update.
- The AI agent waits indefinitely instead of continuing with its next planned tool execution.
- Log files indicate "finish_reason=stop" on helper text that does not actually request user input.
- Developers need to automate the detection and regression-testing of control-loop stalls in WSL or local environments.
How Hermes Continuation Maintainer Works
- Scanning: Runs a diagnostic script over recent logs (e.g.,
agent.log) and state databases (state.db) to identify potential premature stops. - Classification: Categorizes findings into actionable groups such as
continuation_candidate,asks_user, orfinal_or_summary. - Targeted Patching: Applies deterministic heuristic updates to specific control-loop files (
agent_runtime_helpers.py,conversation_loop.py,run_agent.py). - Regression Testing: Automatically executes python pytest commands to verify that the change fixes the stall without breaking user input requests or final summaries.
- Reporting: Outputs a clean, concise markdown report outlining root causes, scanner counts, modified files, and process restart statuses.
Hermes Continuation Maintainer Setup
To begin diagnostics, execute the bundled scanner from the Hermes WSL directory:
python3 <skill_dir>/scripts/scan_hermes_continuation.py --since-minutes 360 --format text
To apply and verify regression checks on the agent environment:
cd /usr/local/lib/hermes-agent
PYTHONDONTWRITEBYTECODE=1 venv/bin/python -m pytest -q \
tests/run_agent/test_run_agent.py::test_run_conversation_continues_after_nonfinal_work_status_after_tools \
tests/run_agent/test_run_agent.py::test_run_conversation_continues_after_chinese_future_action_status \
tests/run_agent/test_run_agent.py::test_run_conversation_does_not_continue_when_status_requests_user_input
Hermes Continuation Maintainer Data Schema & Taxonomy
The skill interacts with the following log files and databases to assess status and apply classification logic:
| Path | Type | Description |
|---|---|---|
/root/.hermes/logs/agent.log |
Log File | Contains sequential runtime details used by the scanner to track finish_reason=stop events. |
/root/.hermes/state.db |
SQLite DB | Stores state variables and conversation loop history for agent execution. |
/usr/local/lib/hermes-agent/ |
Root Directory | The base path for editing Python control-loop modules. |
Classification Taxonomy
continuation_candidate: The agent stopped prematurely and should be patched.asks_user: A valid stop where the agent requested input.final_or_summary: A valid stop indicating complete task execution.review: Ambiguous log state needing manual developer review.
Hermes Continuation Maintainer Advanced Features
- Deterministic Log Scanning: Allows time-window filtering (using
--since-minutes) to pinpoint recent control-loop failures rapidly. - Heuristic-Driven Logic Patches: Ensures changes to the agent loop require execution-oriented keywords (e.g., run, test, execute, verify) and systematically reject prompts requiring direct user clarification.
- Built-in Regression Suite: Minimizes development risk by running focused, pre-configured pytests across multiple critical test suites.
- Interactive CLI Safeguards: Explicitly warns operators before making changes that require restarting active, non-hot-loaded CLI environments.
SKILL.md
Loading
Related Openclaw Skills
Global Trend Radar
A comprehensive trend intelligence skill that monitors, analyzes, and synthesizes emerging global trends across technology, economics, climate, and innovation.
Tavily Search & Research
An AI-optimized search and extraction skill that connects agents to the web using Tavily's specialized RAG-friendly APIs.
Pronoun Resolver
A zero-latency local context hook that detects ambiguous pronouns, vague referents, and bare imperatives, helping AI agents resolve them accurately based on conversation history.
Exa Search
A specialized semantic search, clean web scraping, and content extraction skill engineered specifically for AI agents and LLMs.
ZeroHub Image API
An advanced Openclaw Skills integration for zeroHub, enabling secure, automated text-to-image and image-to-image workflows directly inside your AI agent environment.
SEO Blog Writer
An end-to-end editorial pipeline that transforms single long-tail search queries into search-optimized, structured blog posts across multiple CMS targets.
Featured*
