메뉴
Manage ResourceX resources, register prototypes, and load skills. Use when you need to add, search, distribute, or inspect resources in RoleX, or when you need to register a prototype or load a skill on demand.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Manage individual lifecycle and knowledge injection. Use when you need to create, retire, restore, or permanently remove individuals, or when you need to inject principles and procedures into an individual.
Manage issues between individuals using IssueX. Use when you need to publish issues, comment, close, assign, or label issues for structured communication between AI individuals.
Manage organizations and positions — founding, chartering, membership, duties, requirements, appointments, and dissolution.
Author prototype packages — create the directory structure, prototype.json instruction set, and .feature content files. Use when creating a new role or organization prototype from scratch.
Manage prototypes — settle and evict. Use when you need to register or remove prototypes from the runtime.
Guide for creating RoleX skills. Use when creating a new skill or updating an existing skill that extends a role's capabilities. Skills are SKILL.md files referenced by a procedure — the procedure summary is loaded at activate time, the full SKILL.md is loaded on demand via skill(locator).
| name | resource-management |
| description | Manage ResourceX resources, register prototypes, and load skills. Use when you need to add, search, distribute, or inspect resources in RoleX, or when you need to register a prototype or load a skill on demand. |
Feature: ResourceX Concepts ResourceX is the resource system that powers RoleX's content management. Resources are typed content bundles identified by tag and digest, stored locally in a CAS (Content-Addressable Storage) or in a remote registry.
Scenario: What is a resource Given a resource is a directory containing content and metadata Then it has a resource.json manifest defining name, type, tag, and author And it contains content files specific to its type (e.g. .feature files, SKILL.md) And it is identified by a locator string
Scenario: Tag + digest model Given ResourceX uses a tag + digest model similar to Docker Then tag is a human-readable label — like "stable" or "0.1.0" And tag is mutable — the same tag can point to different content over time And digest is a sha256 content fingerprint — deterministic and immutable And digest is computed from the archive's file-level hashes And the format is sha256:<64-char-hex> And content uniqueness is guaranteed by digest, not by tag
Scenario: Locator formats Given a locator is how you reference a resource Then it can be just a name — nuwa (tag defaults to latest) And it can be a name with digest — name@sha256:abc123... And it can include a path — path/name:tag (e.g. prompts/hello:stable) And it can include a registry — registry.example.com/name:tag And it can be a local directory path — ./path/to/resource or /absolute/path And when tag is omitted, it defaults to latest
Scenario: Resource types in RoleX Given RoleX registers resource types with ResourceX Then "role" type — individual manifests with .feature files (alias: "individual") And "organization" type — organization manifests with .feature files (alias: "org") And "skill" type — SKILL.md files loaded via the skill process
Scenario: resource.json structure Given every resource directory must contain a resource.json Then the structure is: """ { "name": "my-resource", "type": "role", "tag": "0.1.0", "author": "deepractice", "description": "What this resource is" } """ And name is the resource identifier And type determines how the resource is resolved (role, organization, skill) And tag is a human-readable label, not a semantic version — omit for latest
Scenario: Storage location Given resources are stored locally at ~/.deepractice/resourcex by default And the location is configurable via LocalPlatform resourceDir option And prototype registrations are stored at ~/.deepractice/rolex/prototype.json
Feature: Resource Operations Manage resources through the use tool with !resource namespace. Operations cover the full lifecycle: add, push, pull, search, remove.
Scenario: add — import a resource from a local directory Given you have a resource directory with resource.json When you call use with !resource.add Then the resource is archived and stored in local CAS And it gets a digest computed from its content And it can then be pushed to a remote registry And parameters are: """ locator: "!resource.add" args: path: "/absolute/path/to/resource" """
Scenario: push — publish a resource to a remote registry Given a resource has been added to local CAS When you call use with !resource.push Then the resource archive is uploaded to the configured registry And the registry stores it by name, tag, and digest And parameters are: """ locator: "!resource.push" args: locator: "name:tag" registry: "https://..." # optional """
Scenario: pull — download a resource from a remote registry Given a resource exists in a remote registry When you call use with !resource.pull Then the resource is downloaded and cached in local CAS And subsequent resolves use the local cache And parameters are: """ locator: "!resource.pull" args: locator: "name:tag" """
Scenario: search — find resources in local CAS Given you want to find resources stored locally When you call use with !resource.search Then matching resources are returned as locator strings And parameters are: """ locator: "!resource.search" args: query: "keyword" """
Scenario: has — check if a resource exists locally Given you want to verify a resource is in local CAS When you call use with !resource.has Then returns whether the resource exists And parameters are: """ locator: "!resource.has" args: locator: "name:tag" """
Scenario: remove — delete a resource from local CAS Given you want to remove a resource from local storage When you call use with !resource.remove Then the resource manifest is removed from local CAS And parameters are: """ locator: "!resource.remove" args: locator: "name:tag" """
Scenario: Typical workflow — add then push Given you want to publish a resource to a registry Then the sequence is: """ 1. locator: "!resource.add" args: { path: "./my-resource" } 2. locator: "!resource.push" args: { locator: "my-resource" } """ And add imports to local CAS, push uploads to registry And tag defaults to latest when omitted
Feature: Resource Loading via use Load resources on demand through the unified use entry point. The use tool dispatches based on locator format.
Scenario: use — load a ResourceX resource Given you need to load or execute a resource When you call use with a regular locator (no ! prefix) Then the resource is resolved through ResourceX and its content returned And parameters are: """ locator: "hello-prompt" # by registry locator (tag defaults to latest) locator: "./path/to/resource" # by local path """
Scenario: skill — load full skill content by locator Given a role has a procedure referencing a skill via locator When you need the detailed instructions beyond the procedure summary Then call skill with the locator from the procedure's Feature description And the full SKILL.md content is returned with metadata header And this is progressive disclosure layer 2 — on-demand knowledge injection And parameters are: """ skill locator: "skill-creator" # tag defaults to latest skill locator: "/absolute/path/to/skill-directory" """
Scenario: Progressive disclosure — three layers Given RoleX uses progressive disclosure to manage context Then layer 1 — procedure: metadata loaded at activate time (role knows what skills exist) And layer 2 — skill: full instructions loaded on demand via skill(locator) And layer 3 — use: execution of external resources via use(locator) And each layer adds detail only when needed, keeping context lean
Feature: Prototype Registration Register a ResourceX source as a role or organization prototype. Prototypes provide inherited state that merges with an individual's instance state on activation.
Scenario: What is a prototype Given an individual's state has two origins — prototype and instance Then prototype state comes from organizational definitions (read-only) And instance state is created by the individual through execution (mutable) And on activation, both are merged into a virtual combined state
Scenario: Prototype resource structure for a role Given a role prototype is a directory with: """ / ├── resource.json (type: "role") ├── individual.json (manifest with id, type, children tree) ├── .individual.feature (persona Gherkin) └── ..feature (identity, background, duty, etc.) """ And individual.json defines the tree structure: """ { "id": "nuwa", "type": "individual", "alias": ["nvwa"], "children": { "identity": { "type": "identity", "children": { "background": { "type": "background" } } } } } """
Scenario: Prototype updates Given you re-register a prototype with the same id Then the source is overwritten — latest registration wins And the next activation will load the updated prototype
Feature: Common Workflows Typical sequences of operations for resource management.
Scenario: Publish a prototype to registry Given you have a prototype directory ready When you want to make it available via registry Then the sequence is: """ 1. locator: "!resource.add" args: { path: "/path/to/roles/nuwa" } 2. locator: "!resource.push" args: { locator: "nuwa" } """ And the prototype is now pullable by anyone with registry access
Scenario: Update and re-push a prototype Given the prototype content has changed When you re-add and push with the same tag Then the registry updates the tag to point to the new digest """ 1. locator: "!resource.add" args: { path: "/path/to/roles/nuwa" } 2. locator: "!resource.push" args: { locator: "nuwa" } """ And consumers pulling the same tag get the updated content
Scenario: Test loading a skill Given you want to verify a skill is accessible When you call skill with the locator Then the full SKILL.md content should be returned And example: """ skill locator: "skill-creator" """
