Crate kernex_skills 

Skill and project loader for Kernex.

Scans {data_dir}/skills/*/SKILL.md and {data_dir}/projects/*/ROLE.md for definitions and exposes them to the system prompt so the AI knows what tools and contexts are available.

Structs

Permissions

Permission categories that a skill can request.

Project

A loaded project definition.

RiskDetector

High-risk permission patterns that trigger warnings.

RiskWarning

A warning about a risky permission.

Skill

A loaded skill definition.

SkillContext

Output of build_skill_prompt: the prompt text plus an optional model override.

Enums

RiskCategory

Categories of permission risks.

TrustLevel

Trust level assigned to a skill based on its source.

Constants

DEFAULT_TRUSTED_ORGS

Trusted organizations whose skills can be auto-approved.

Functions

build_skill_prompt

Build the skill block appended to the system prompt.

determine_trust_level

Determine trust level based on skill source.

ensure_projects_dir

Create {data_dir}/projects/ if it doesn’t exist.

get_project_instructions

Find a project by name and return its instructions.

get_toolboxes_for_skill

Return the toolboxes declared by a named skill.

lazy_skill_directory

Build a compact listing of available lazy skills for prompt injection.

load_projects

Scan {data_dir}/projects/*/ROLE.md and return all valid projects.

load_skills

Scan {data_dir}/skills/*/SKILL.md and return all valid skill definitions.

match_skill_toolboxes

Match user message against skill triggers and return activated toolboxes.

match_skill_triggers

Match user message against skill triggers and return activated MCP servers.

migrate_flat_skills

Migrate legacy flat skill files ({data_dir}/skills/*.md) to the directory-per-skill layout ({data_dir}/skills/{name}/SKILL.md).

skill_search_toolbox

Return the Toolbox definition for the virtual skill_search tool.