Search Results for "Stonefish-Labs" | AgentCC

agent-design-knowledge-retrieval

Knowledge and retrieval consulting skill for corpus design, retrieval architecture selection, awareness orchestration, and staleness management. Use when designing file indexes, MCP retrieval services, custom RAG pipelines, or long-term knowledgebase maintenance.

new-hook

Scaffold a new standalone hook configuration by collecting parameters and executing the hook scaffolding script. Use this skill when the user explicitly requests to "create a hook", "build a new hook", "scaffold a hook event", or generate a pre-tool/post-tool hook. It prompts the user for the hook event name, type (command, prompt, agent), async flag, and target directory. It generates the required `.json` configuration and shell scripts. Do not use this for generating regular skills, agents, or MCP servers.

agent-design-core

Router and governance skill for the agent-design suite. Use when the request spans multiple agent-design domains, when deciding which domain skill(s) to invoke first, when enforcing suite-wide standards, or when maintaining the living knowledgebase quality gates and coverage controls.

new-command

Scaffold a new slash command markdown file by collecting parameters and executing the scaffolding script. Use this skill when the user wants to "create a slash command", "add a custom command", "build a /command", or "scaffold a command". It walks through collecting the command name, description, scope (project or user), optional tool restrictions, model override, argument hints, namespace, bash context blocks, and file references, then generates a spec-compliant .md file in the correct .claude/commands/ location. Do not use this for creating skills, agents, hooks, or MCP servers.

agent-design-workflow-delegation

Workflow decomposition and delegation consulting skill for deciding monolith vs staged handoffs, specifying subagent contracts, and designing isolation-broker patterns. Use when planning subagent boundaries, context budgeting, review gates, recovery paths, or delegation security rules.

refine-skill

Review and score skills for quality, correctness, and effectiveness. Use this skill whenever you need to evaluate an existing skill, audit a skill before deployment, check a skill for bugs or dead code, or score a skill's quality on a standardized rubric. Also use it when someone asks "is this skill any good?", "review my skill", "rate this skill", "what's wrong with this skill", or "how can I improve this skill". This skill provides both automated static analysis (dead code, unused imports, frontmatter validation) and a structured rubric for subjective evaluation (triggering quality, conciseness, writing style). The output is a scored report with concrete, prioritized fixes. Use this skill even for quick gut-checks — a fast review catches bugs that cost hours to debug later.

new-mcp

Scaffold a new independent MCP server using FastMCP 3.x. Use this skill when the user asks to "create an MCP server", "build a new Model Context Protocol server", or "scaffold an MCP integration". Asks for server name and target directory, then generates a ready-to-run FastMCP 3.x project (server.py, pyproject.toml, .mcp.json). Do not use this for standard Claude skills or agents.

agent-design-output-reliability

Output reliability consulting skill for layering schema constraints, coverage determinism, factual verification, and rubric-based quality gates. Use when designing structured outputs, acceptance checks, quality scoring, or reliability guardrails for agent-generated artifacts.

new-skill

Scaffold a new standalone skill component by collecting parameters and executing the scaffolding script. Use this skill when the user wants to "create a new skill", "build a standalone skill", or "scaffold a skill". It handles collecting all required properties (name, description, target directory) and optional metadata (license, compatibility, author) before generating the strictly-formatted folder structure containing SKILL.md and its progressive disclosure folders (scripts/, assets/, references/). Do not use this for agents or MCP servers.

new-plugin

Act as a Packager to take a list of independent components (skills, agents, hooks, MCP servers) and assemble them into a strictly-formatted Anthropic plugin bundle. Use this skill when the user asks to "package a plugin", "bundle these components", "create a plugin.json", or "deploy a plugin". It collects the target components and an output directory, then generates the `.claude-plugin/` structure and `plugin.json` manifest. Do not use this to create new components from scratch.

agent-design-tooling-security

Tooling and security consulting skill for selecting native tools, hooks, MCP, or scripts with explicit security boundaries and secrets strategy. Use when designing agent tool interfaces, permission models, consent gating, or durable configuration and secret handling.

agent-design-architecture

Architecture consulting skill for choosing and evolving agent system topology across programmatic, assistant, and autonomous models. Use when deciding orchestration authority, termination logic, control surfaces, migration paths, or architecture anti-pattern mitigations.

publish

Publish a local plugin, plugin bundle, or standalone component directory to a GitHub repository. Handles git initialization, topic tagging, and marketplace entry generation. Use when the user wants to "publish", "push to GitHub", "release", or "upload" a plugin, skill, agent, hook, MCP server, or command to a remote repository.

validate-plugin

Validate a plugin directory for structural correctness, manifest compliance, and component validity. Use when checking if an Anthropic plugin is ready to publish or run locally.

new-agent

Scaffold a new independent agent component by collecting parameters and executing the agent scaffolding script. Use this skill when the user asks to "create an agent", "build a new agent", or "generate an independent agent". It walks the user through collecting the agent name, target directory, scope, isolation mode, and background flag, then builds a spec-compliant single markdown file for the agent. Do not use this for generating skills, hooks, or MCP servers.