Loading ToolBench

Reading the current corpus…

haiku.skills MCP server · F · ToolBench

← Browse

haiku.skills

github.com/ggozad/haiku.skills

Skill-powered AI agents implementing the Agent Skills specification with pydantic-ai

2026-07-28+STDIO4 tools

47/100

Definition Quality42

Protocol Readiness50

Supportability52

Spec posture

Inferred effective spec 2026-07-28+·current baseline 2026-07-28

Static source inference · medium confidence

Current-spec patterns detected

Summary

This server has critical definition quality gaps. Only 2 of 4 tools have non-trivial descriptions. Parameter descriptions are completely absent (all marked null in schema). Tools lack proper composition, 'run_script' is a generic execution wrapper that violates single-responsibility principles. No error handling guidance visible. No output schemas documented. This server would not pass code review for production use.

Tools (4)

addread onlysource verified38/100

Add two numbers.

greetread onlysource verified40/100

Greet someone by name.

read_resourceread onlysource verified52/100

Read a resource file from the skill directory.

run_scriptirreversiblesource verified37/100

Execute a script from the skill's scripts/ directory.

Issues (6)

CRITICALtool-description ↗

All parameter descriptions are null. The schema shows {'a': {'type': 'integer', 'description': null}, 'b': {'type': 'integer', 'description': null}} for 'add', and similar patterns for 'greet', 'read_resource', and 'run_script'. LLMs cannot infer parameter meaning without descriptions.

addgreetread_resourcerun_script

CRITICALtool ↗

'run_script' is a generic execution tool that accepts arbitrary scripts and arguments. This violates single-responsibility principles and is a security risk. It should either be removed or replaced with domain-specific tools (e.g., 'execute_analysis_script', 'execute_extraction_script') with fixed allowed scripts.

run_script

HIGHtool ↗

No output schemas are documented for any tool. Tool descriptions do not state what fields are returned or what structure the response has. LLMs cannot plan downstream operations without knowing the shape of returned data.

addgreetread_resourcerun_script

Recommendations

Add non-null descriptions to ALL parameters. For 'add', change {'a': {'type': 'integer', 'description': null}} to {'a': {'type': 'integer', 'description': 'The first integer to add'}}. Apply to all 4 tools.
Document output schemas for each tool in the tool description. E.g., 'Returns: {"result": integer, "timestamp": string}' for 'add'. Make the response structure explicit so agents know what fields to expect.
Replace or remove 'run_script'. If scripts are needed, replace with domain-specific tools: 'execute_extraction_script(format: enum[csv|json|xml])' with a fixed, audited list of available scripts. If keeping it, enforce a whitelist: validate 'script' against allowed_scripts=['extract.py', 'transform.py'], and validate 'arguments' with a strict regex/schema.
Expand descriptions for 'add' and 'greet' to 50 - 200 characters explaining when an agent should use them. E.g., 'Add two integers. Use this tool when the user asks for arithmetic operations or when summing numeric values in downstream steps.'
Add path constraints to 'read_resource': 'Relative path to a resource file in the skill directory (must not contain \'../\' or absolute paths). Example: \'data/config.yaml\'.'
Add error handling guidance. Document what happens if a file is not found, a script fails, or an invalid path is passed. E.g., 'Returns error: "File not found. Available resources: [list]. Did you mean one of these?"'
Validate all inputs server-side. For 'run_script', check that 'script' is in a whitelist and 'arguments' matches a safe pattern before execution. Return 400 Bad Request with actionable error message (not a 500 stack trace).

Score history

Overall score trend

↑ 21 points across a rubric change (v1 → v2)

47/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-22	F	47	2026-07-28+	v2
2026-03-09	F	26	-	v1