Help us improve
Share bugs, ideas, or general feedback.
From playbooks-virtuoso
Helps discover the right skill, agent, or team for a task by scanning installed components at runtime and matching the user's situation.
npx claudepluginhub krzysztofsurdy/code-virtuoso --plugin agents-virtuosoHow this skill is triggered — by the user, by Claude, or both
Slash command
/playbooks-virtuoso:using-ecosystem [optional: situation or topic][optional: situation or topic]The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Discover the right skill, agent, or team for the situation at hand. This skill teaches a discovery PROCESS - it does not maintain a hardcoded inventory. Every recommendation comes from scanning what is actually installed.
Recommends the right agents (Task tool) and skills (Skill tool) for any task. Helps users find and choose between heavyweight agents and lightweight skills.
Discovers and invokes agent skills for engineering workflows. Helps identify and apply the right skill for the current development phase or task.
Interviews users to clarify their goal and recommends the best installed skills for the task. Use when a user is unsure which skill to invoke or how to start.
Share bugs, ideas, or general feedback.
Discover the right skill, agent, or team for the situation at hand. This skill teaches a discovery PROCESS - it does not maintain a hardcoded inventory. Every recommendation comes from scanning what is actually installed.
| Principle | Meaning |
|---|---|
| Scan, don't guess | Always scan installed skills/agents/teams before recommending. Never assume something is available. |
| Match situation, not topic | Pick the skill whose description triggers match what the user is doing, not the skill whose title loosely matches. |
| Agents act, skills inform | Skills are reference material. Agents are actors. Delegate work to an agent; consult a skill. |
| Teams for multi-role work | If the work needs 3+ agents coordinating, check for a pre-composed team before assembling ad-hoc. |
| Narrower wins | If two skills overlap, pick the one with the tighter trigger match. |
| Chain when work crosses domains | Multi-step tasks often need two or three agents in sequence. |
Run the discovery commands from discovery-commands to build a live index of installed skills, agents, and teams. Read each frontmatter name and description field.
Do NOT skip this step. Do NOT recommend from memory. The installed set varies per user and per project.
Determine what the user needs:
| Signal | Category |
|---|---|
| Needs reference material, patterns, or principles | Knowledge skill |
| Needs to generate a specific output (ticket, PR message, report, rules file) | Tool skill |
| Needs to follow a step-by-step operational procedure | Playbook skill |
| Needs framework-specific component reference | Framework skill |
| Needs a single focused task done (investigate, review, implement) | Specialist agent |
| Needs ongoing domain ownership (requirements, architecture, QA) | Role agent |
| Needs coordinated multi-agent delivery (feature, release, review cycle) | Team |
User request
|
v
Does it need multiple agents coordinating?
+-- Yes --> Scan bundled and project teams for a match
| +-- Found --> Recommend the team
| +-- None --> Assemble ad-hoc from agents (see chaining below)
|
+-- No --> Is it a well-scoped actionable task?
+-- Yes --> Scan agents/ for a matching specialist or role
+-- No --> Scan skills/ for a matching skill
Use the category signals from Step 2
description field of each candidate. Match on trigger phrases, not the name.database-design beats clean-architecture for a schema question.user-invocable - if the user wants to run something interactively, only recommend user-invocable skills.security + solid for a security review).npx skills add.When a single skill or agent is not enough, chain them. Common patterns:
| Pattern | Shape |
|---|---|
| Investigation flow | Read-only agent --> Design agent --> Implementation agent --> Review agent |
| Feature flow | Requirements role --> Architecture role --> Dev roles (parallel) --> QA role |
| Review flow | Smell scanner --> Reviewer --> (optional) Implementer for fixes |
| Coverage flow | Test gap analyzer --> Implementer --> Reviewer |
These are guidelines, not rituals. Skip steps that do not apply. Add steps that do. If a pre-composed team matches the chain, use the team instead.
When the user has just finished work with an agent, skill, or team and asks "what next?" or "where do I go from here?":
Compass mode is conversational. Ask the user, do not scan the filesystem to infer what happened.
The ecosystem has five layers. The structure is stable - it does not change when skills are added.
| Layer | Location | Format | Purpose |
|---|---|---|---|
| Skills | skills/{category}/{name}/SKILL.md | Markdown + YAML frontmatter | Reference material, tools, playbooks, roles, frameworks |
| Agents | agents/{name}.md | Markdown + YAML frontmatter | Specialist and role sub-agents |
| Teams | Bundled inside dispatching skill or at project teams/{name}.md | Markdown + YAML frontmatter | Pre-composed agent teams with coordination protocols |
| Specs | spec/*.md | Markdown | Format specifications for skills, agents, teams, plugins |
| Plugins | .claude-plugin/marketplace.json | JSON | Distribution bundles grouping skills and agents |
| Category | Path | Invocation |
|---|---|---|
| Knowledge | skills/knowledge/ | Auto-loaded when relevant |
| Tools | skills/tools/ | User-invocable slash commands |
| Playbooks | skills/playbooks/ | User-invocable slash commands |
| Roles | skills/roles/ | Auto-loaded by matching role agents |
| Frameworks | skills/frameworks/{name}/ | Loaded when the framework is in play |
| Tier | Scope | Examples |
|---|---|---|
| Specialist | Single repeatable task (stateless) | investigator, reviewer, refactor-scout |
| Role | Domain of responsibility (some carry memory) | architect, backend-dev, qa-engineer |
| Mode | When | How |
|---|---|---|
| Peer | Platform supports agent-to-agent messaging | Lead creates team, teammates claim tasks and message each other |
| Sequential | All other platforms | Lead dispatches one sub-agent per phase, passes outputs forward |
Before answering a "which skill/agent/team" question:
| Reference | Contents |
|---|---|
| discovery-commands | Shell commands to scan installed skills, agents, and teams at runtime |
| chaining-patterns | Multi-agent chain templates with when-to-use guidance |
| compass | "What next?" recommendation map for post-task navigation |