PowerShell-based skills and MCP (Model Context Protocol) management system for discovering, installing, managing, and auditing skills and MCP tools
Critical gaps across all dimensions. Tool names lack clear action verbs (audit-targets, capability-inventory, rule-audit use hyphens inconsistently; many are compound nouns rather than verb_noun patterns). Descriptions are present but generic (10-50 chars, well below the 50-200 char LLM-optimized baseline). Parameter schemas exist but are minimal: most tools accept only 1-3 params (json, plan, mode flags), with no enums, ranges, or validation constraints. No output schemas documented. Error handling absent from visible code. Security concerns: no evidence of permission gates, audit logging, or secret injection patterns. Tool composition shows poor separation of concerns (audit-targets combines dry-run planning AND apply execution; mcp-sync does configuration AND state synchronization). No pagination, no idempotency markers, no recovery guidance.
Audit and plan skill/MCP target state changes with dry-run and apply modes
Generate inventory of available capabilities from skills and MCP sources
Check for available updates to skills and MCP tools
Diagnose system health and configuration issues
Apply global rule changes
Check global rules compliance
Plan global rule changes
Tool names lack clear action verbs. Names like 'audit-targets', 'capability-inventory', 'rule-audit' are noun-heavy and use hyphens inconsistently. LLMs cannot infer intent from compound nouns. Rename to verb_noun pattern: 'audit_targets', 'list_capabilities', 'check_rules'.
Descriptions are generic and under 50 characters (e.g., 'Audit and plan skill/MCP target state changes with dry-run and apply modes' is 75 chars but lacks WHEN/WHY context). Baseline for LLM-optimized descriptions is 50-200 chars with explicit action, use case, and prerequisites. Rewrite all descriptions to answer: What does it do? When should the LLM call it? What does it return?
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 39 | 2026-07-28+ | v2 |
Rollback global rule changes
Install MCP (Model Context Protocol) tools
Manage MCP tool profiles and configurations
Synchronize MCP tool configuration and state
Uninstall MCP tools
Remove invalid skill/MCP mappings
Perform a release update of skills and MCP tools
Schedule automatic release updates
Apply planned rule changes
Audit rules against current state and generate compliance report
Apply planned rule changes across the entire estate
Audit rules across the entire estate/organization
Plan rule changes across the entire estate
Rollback previously applied rule changes across the estate
Plan rule changes for a specific target
Verify lock file integrity and consistency
Parameter schemas are minimal and lack constraints. Most tools accept only boolean flags (json, plan, strict) or bare strings (mode, name, action, schedule) with no enums, ranges, or validation rules. E.g., 'mode' in audit-targets accepts 'dry-run' or 'apply' but this is not declared as an enum. Add JSON Schema enums, patterns, and descriptions for all parameters.
No output schemas documented. Code shows complex JSON structures (e.g., New-AuditDryRunSummary returns ordered objects with nested arrays, counts, manifests) but no schema is exposed to the LLM. LLMs cannot plan downstream calls or extract fields without knowing the response structure. Document all output schemas in tool definitions.
No error handling guidance visible. Code does not show recovery hints, error categorization (retryable vs fatal), or actionable error messages. Agents cannot self-correct without explicit guidance. Add error responses that tell the LLM what to do next and classify errors as retryable, user-fixable, or fatal.