ASI agent-owned CLI providing deterministic skills boundary with support for creator and onboard skills
This is a CLI skill toolkit, not an MCP server. The tools are shell script wrappers around a Python CLI (`asi`), not MCP-registered tools with schemas. Tool definitions are inferred from command-line interfaces rather than explicitly registered via MCP protocol. No JSON Schema input/output definitions are visible in the source. Descriptions are minimal (under 20 chars for most tools). Parameter schemas exist only for some tools (e.g., 'skill init', 'creator migrate') as command-line flags, not as structured JSON Schema. This server fundamentally misaligns with the MCP tool definition model: it provides shell command wrappers, not remotely-callable tools with documented parameters, output schemas, or error handling guidance.
Creator skill
Apply user answers
Migrate legacy artifacts
Emit next questions
Run creator via session goal
Validate agent suggestions
Verify dependencies
Onboard skill via plan JSON
Execute onboard via plan JSON
No MCP tool registration found. Tools are inferred from CLI wrappers, not explicitly registered via MCP protocol with JSON Schema definitions. This violates the core MCP contract.
Descriptions are too short and non-descriptive. Most tools have descriptions under 30 characters (e.g., 'Creator skill', 'Emit next questions'). These do not explain WHAT the tool does, WHEN to use it, or WHAT prerequisites exist. LLMs cannot determine when or why to select these tools.
Input schemas are missing or incomplete. Only 3 tools have any visible schema structure: 'skill init' (skill_dir string), 'onboard' and 'onboard run' (plan object), 'creator suggest' and 'creator apply' (object input), 'creator migrate' (source enum + force boolean). Most tools (doctor, creator, creator next) have no visible input schema at all. No output schemas are documented anywhere.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 19 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 28 | - | v1 |
Emit concatenated references
No parameter descriptions. Where schemas exist, parameter descriptions are minimal (e.g., 'Directory path to the skill', 'Onboarding plan from stdin'). Parameters lack context about format, constraints, or valid values. The 'plan' parameter appears in multiple tools but is never fully documented as to structure or required fields.
No error handling guidance. Tools provide no recovery hints, error classification (retryable vs fatal), or actionable error messages. If a tool fails, the LLM has no guidance on what to do next.
Vague naming. Tools like 'doctor', 'creator', 'onboard' do not start with clear action verbs. 'doctor' could mean diagnose, validate, or check, unclear. 'creator' is a noun, not a verb. These names do not convey what happens when called. Better names: 'verify_dependencies', 'init_skill_creator', 'apply_onboarding_plan'.
Multiple tools share responsibility without clear distinction. 'onboard' and 'onboard run' both accept a 'plan' object and appear to do the same thing (execute onboarding). Similarly, 'creator run', 'creator suggest', 'creator apply', and 'creator next' form a multi-step interaction, but the distinct roles and sequencing are not documented. This forces LLMs to guess which tool to call.
No output schema documentation. It is unclear what each tool returns. 'skill init' says 'Emit concatenated references', what does that look like? 'creator next' says 'Emit next questions', is that JSON, plain text, or structured? Output schemas are required so LLMs can plan downstream tool calls.