Skills

Reusable, versioned workflow instructions for agents, managed by the Nexus platform.

Skills provide structured guidance for agents — from session initialization to security audits. They are loaded live from the Nexus backend at invocation time and exposed as slash commands in the agent runtime.

What is a skill?

A skill is a markdown-based workflow instruction assigned to a project. Each skill defines:

• A purpose — what the skill does
• A workflow — step-by-step instructions for the agent
• Rules — constraints and guardrails
• Validation — checks to verify the skill was executed correctly

Skills are versioned, searchable, and exposed to agents as slash commands (e.g. /nexus-init, /nexus-adr-draft). When an agent invokes a skill command, the skill body is loaded live from the Nexus backend — not from local files.

How skills work

1. Assignment — skills are assigned to projects in the Nexus dashboard or via MCP tools
2. Sync — nexus pull writes the skill files (with their resources) to the agentic root (.nexus/skills/) and, for Claude Code projects, to .claude/skills/<skill>/
3. Invocation — the agent invokes a slash command (e.g. /nexus-sec-scan)
4. Loading — the skill body is fetched live from the backend via the MCP server
5. Execution — the agent follows the skill's instructions

Built-in skills

Skill	Command	Purpose
Init Nexus	/nexus-init	Initialize a session, load project context, sweep dispatches
Session Close	/nexus-session-close	Close the active session with a structured summary and next entry point
ADR Draft	/nexus-adr-draft	Draft an Architecture Decision Record with context, decision, consequences
Sync Memory	/nexus-sync-memory	Classify and persist session outputs to the knowledge base
Grill Me	/nexus-grill-me	Structured grilling session to stress-test a plan, design, or architecture before implementation
Deploy Check	/nexus-deploy-check	Pre-deployment verification: build, lint, type-check, test, environment validation
DB Migrate	/nexus-db-migrate	Generate and validate database migration scripts for Supabase/PostgreSQL
Code Review	/nexus-code-review	Structured code review with checklist-based analysis
Security Scan	/nexus-sec-scan	OWASP Top 10 security audit with structured findings report
Performance Scan	/nexus-perf-scan	Performance audit using Lighthouse (web) or profiling tools (CLI)
Learnings	/nexus-learnings	Capture surprises, unexpected findings, and gotchas as research notes
Dispatch Sweep	/nexus-dispatch-sweep	Process open Dispatches: prioritized sweep, acknowledge, act by type, reply, resolve and close
Dispatch Create	/nexus-dispatch-create	Create a structured Dispatch for agent coordination
Show Plugins	/nexus-show-plugins	Display loaded Nexus plugins and connection status

Catalog cleanup (v0.16.0)

• Archived skills were removed from the catalog: nx-sec-check, nx-perf-check, nx-sprint-plan, nx-test-runner, nx-research-digest and nx-init-nexus.
• nx-vl-sweep was merged into nx-dispatch-sweep (/nexus-dispatch-sweep); its project assignments moved over.
• nx-cr-review (CodeRabbit findings) is a short skill with four resource files.
• nx-grill-me has a real description and runs only on request.

Which skill owns which workflow, and which plugins or built-in commands must not replace it, is written into each project's capability routing rule.

Descriptions

Every skill needs a description of at most 500 characters. Agents pick skills by this line, so it should say what the skill does and when to use it. Create and update reject a missing or blank description, in MCP (sk_create, sk_update) and REST alike.

Run only on request

Some skills should run only when the user asks for them, never because the model decided to. Switch on Run only on request in the skill form, or set invocation in sk_create / sk_update:

invocation	Behavior
model (default)	The agent may start the skill on its own when it fits the task.
user	The skill runs only when the user invokes it.

For Claude Code, nexus pull (nexus-cli 0.28.9 or later) renders invocation: user as disable-model-invocation: true in the skill's SKILL.md frontmatter. OpenCode has no equivalent; its command stub is unchanged.

Skill resources

Multi-file skills ship resource files (references, policies, examples) next to the skill body. sk_export returns them, and nexus pull writes them into the skill's folder (.claude/skills/<skill>/ and the agentic root). Before v0.16.0, resource files of skills such as nx-cr-review or nx-grill-me never reached the workspace.

Skill scope

• Core skills (project_id = NULL) are tenant-wide and available to all projects. Only platform owners and admins can create or modify them.
• Project-scoped skills belong to a single project and are managed by project admins and editors. They do not appear in other projects unless explicitly promoted to core.

Permissions (v0.16.0)

Tool	Core skill	Project skill
sk_create	platform admin (sk_create always creates a core skill)	n/a
sk_update, sk_activate, sk_delete	platform admin	project admin or editor
sk_assign, sk_unassign	project admin or editor	project admin or editor

MCP tools

Tool	Description
sk_list	List skills for the current tenant
sk_get	Get full skill content by identifier
sk_create	Create a new skill in draft status (description required, optional invocation)
sk_update	Update skill content (increments version; optional invocation)
sk_activate	Change skill status (draft, active, archived)
sk_assign	Assign a skill to a project
sk_unassign	Remove a skill assignment from a project
sk_export	Export all skill assignments for a project, including invocation and resource files