Tool Repair Skill for Hermes and OpenCode

A harness-level fix for LLM tool calling. Catches the common JSON formatting mistakes open models make and fixes them deterministically before the tool executor ever sees them. Ships with adapters for three agent frameworks:

Based on the approach that made DeepSeek V4 Pro outperform Opus 4.7 on tool calling (see CommandCode's post and YouTube deep dive).

The Problem

Open models (DeepSeek, GLM, Qwen, Kimi) make the same tiny JSON mistakes in tool calls over and over. Each mistake triggers a validation error. The model retries with the same bad format. The session degrades through 50+ wasted retry cycles. The model never learns because the error messages are opaque.

These mistakes are not random. They are a small finite set of patterns caused by the model's training distribution leaking through the tool boundary.

Harness vs Model

Most people frame this as a model problem: "DeepSeek is bad at tool calling, wait for the next version." That is wrong. It is a harness problem. The harness sits between the model and the tool executor. It decides what to do with the model's output: reject it and waste tokens retrying, or fix it silently and move on. A harness that repairs deterministically turns a bad-at-tool-calling model into a functional one in about 200 lines of code.

The model did not change. The harness got more forgiving in exactly the places it needed to be.

The Four Patterns This Fixes

How It Works

flowchart TD
    subgraph Harness["HARNESS BOUNDARY"]
        direction TB
        P["Parse JSON"] --> V{"Schema Valid?"}
        V -->|"Yes"| D["Execute Tool"]
        V -->|"No"| W["Walk Issue List by Path"]
        W --> R["Apply Repairs<br/>in Priority Order"]
        R --> RV{"Re-validate"}
        RV -->|"Pass"| D
        RV -->|"Fail"| E["Return Readable Error<br/>with Guidance"]
    end

    M["Model Output<br/>(raw tool call JSON)"] --> P
    D --> N["Tool Result<br/>+ Repair Note"]
    E --> N
    N --> B["Back to Model"]

Everything inside the HARNESS BOUNDARY box is your agent framework. The model provides the raw JSON and receives the result. All repair logic, validation, and correction notes are handled at the harness layer.

Key design rule: Valid inputs are never touched. The repair layer parses the input as-is first. If it passes the schema, it ships immediately. Repairs only fire at paths the validator actually flagged. This prevents silent corruption of legitimate data (for example, writeFile content that happens to be JSON-shaped).

Components

tool_repair.py (the core library)

Standalone Python module with no dependencies beyond stdlib. Main entry point:

from agent.tool_repair import repair_function_args

repaired_args, repair_notes = repair_function_args(
    function_name="readFile",
    function_args={"path": "/tmp/test.txt", "limit": None},
    tool_schema=None,  # optional JSON schema for type-aware repairs
)
# repaired_args = {"path": "/tmp/test.txt"}
# repair_notes = ["[repair: null values removed for optional fields]"]

Can be imported and used by any agent framework, not just Hermes.

Hermes Agent integration (included)

Two small modifications to the Hermes harness core. Both operate at the harness layer, between the model's output and the tool executor:

1. agent/agent_runtime_helpers.py. sanitize_tool_call_arguments() is a harness function that walks tool calls before dispatch. It used to only catch unparseable JSON and replace it with {}. Now after json.loads() succeeds, it runs repair_function_args() on the parsed dict. If repairs trigger, it updates the arguments JSON and stores a repair note in the harness side-channel.

2. agent/tool_dispatch_helpers.py. make_tool_result_message() is a harness function that builds the tool result before it goes back to the model. It checks the harness side-channel for pending repair notes and appends them to the result content.

The model reads the repair note alongside the successful result and adapts on the next turn. The harness did the fixing. The model just benefits from seeing what was fixed.

Hermes Plugin (draft)

references/plugin.yaml plus plugin-architecture.md. A blueprint for packaging the repair logic as a proper Hermes plugin with telemetry, dashboard, and config. Needs a pre_tool_call hook that supports argument modification (not currently available in Hermes hook system).

Adapted For Other Frameworks

This repo ships adapters for two other agent frameworks in the adapters/ directory. Each adapter wraps the same core tool_repair.py library with the harness-specific wiring.

OpenCode has the cleanest integration because its tool.execute.before hook supports argument mutation. Claude Code is the most limited. PreToolUse can only block, not mutate, so it wastes a turn when it detects a pattern.

See each adapter's README for setup instructions.

Safety Guarantees

• Valid inputs are never touched. The first step is always "try the input as-is." Only paths that fail validation get repaired.
• Non-JSON tool data is unaffected. The repair layer only examines tool call arguments (the JSON dict describing what the tool should do), not tool results, binary content, images, or multimodal data.
• Schema-aware array repairs. Array-specific repairs (empty-object-to-array, bare-string-wrap) only fire when the tool JSON schema confirms the field expects an array type. Without a schema, only safe universal repairs run (null-strip, stringified-array-parse, autolink-unwrap).
• Repair notes deduplicate. If a repair note was already appended on a previous turn, it won't get stacked again.

Dependencies

The core library (tool_repair.py) needs nothing beyond Python standard library.

No pip packages, no npm modules, no external services for the core library.

How to Install

Core library (any framework)

cp references/tool_repair.py /your/project/tool_repair.py

from tool_repair import repair_function_args
fixed, notes = repair_function_args("my_tool", {"some_field": None})

Hermes Agent

Copy the library and apply the two patches described in Components:

cp references/tool_repair.py /path/to/hermes/agent/tool_repair.py

Or prompt your agent:

Clone https://github.com/bojansandhaus/tool-repair-skill-for-hermes-and-opencode.git, copy references/tool_repair.py into the Hermes agent directory, and enable agent.tool_repair: true in ~/.hermes/config.yaml.

Enable in ~/.hermes/config.yaml:

agent:
  tool_repair: true

OpenCode

Copy the TypeScript adapter into your OpenCode plugins directory:

cp -r adapters/opencode/* ~/.config/opencode/plugins/

Or prompt your agent:

Clone https://github.com/bojansandhaus/tool-repair-skill-for-hermes-and-opencode.git and copy the TypeScript plugin from adapters/opencode/ to ~/.config/opencode/plugins/.

Claude Code

Copy the hook scripts and configure in claude.json:

cp adapters/claude-code/*.sh .claude/hooks/
chmod +x .claude/hooks/*.sh

Or prompt your agent:

Clone https://github.com/bojansandhaus/tool-repair-skill-for-hermes-and-opencode.git, copy the hook scripts from adapters/claude-code/ to .claude/hooks/, make them executable, and add the pre_tool_use and post_tool_use hook entries to claude.json.

{
  "hooks": {
    "pre_tool_use": {
      "matcher": "*",
      "command": "bash .claude/hooks/pre_tool_use.sh"
    },
    "post_tool_use": {
      "matcher": "*",
      "command": "bash .claude/hooks/post_tool_use.sh"
    }
  }
}

Clone the repo

git clone https://github.com/bojansandhaus/tool-repair-skill-for-hermes-and-opencode.git
cd tool-repair-skill-for-hermes-and-opencode

Usage

From any Python project

import json
from tool_repair import repair_function_args

def dispatch_tool(name, args_json):
    args = json.loads(args_json)
    if isinstance(args, dict):
        fixed_args, notes = repair_function_args(name, args)
        if notes:
            print(f"Repaired {name}: {notes}")
            args_json = json.dumps(fixed_args)
    # proceed with the tool call

In Hermes Agent

Already wired in. No additional setup needed. The integration lives in sanitize_tool_call_arguments and make_tool_result_message.

Roadmap

• [x] Core repair library (5 pattern fixes)
• [x] Hermes integration (sanitize + tool result pipeline)
• [x] Repair note side channel (model self-correction)
• [x] OpenCode adapter (TypeScript plugin)
• [x] Claude Code adapter (bash + jq hooks)
• [ ] Schema-aware repairs (type inference from JSON schema)
• [ ] Per-model repair telemetry (dashboard tab)
• [ ] Model-specific repair profiles (DeepSeek, GLM, Kimi quirks)

License

MIT. Free to use, modify, and distribute. This is a direct implementation of patterns discovered by the CommandCode team. Credit for the original insight goes to them.