andrescevp/ia-agents-development
Multi-platform skills for creating, updating, and conciliating AI agent plugins across Claude Code, Codex/ChatGPT, Hermes Agent, OpenCode, and Google Antigravity.
Use this skill when one AI agent plugin must be created, audited, or reconciled across Claude Code, OpenAI Codex/ChatGPT, Hermes Agent, OpenCode V2, and Google Antigravity (plus skill-only hosts such as Copilot CLI). It maps manifests, skills, commands, subagents, hooks, MCP servers, tools, secrets, UI, and distribution across platforms; detects gaps and drift with a compatibility matrix and drift ledger; decides between unify, generate, adapt, degrade, declare, and drop; and validates cross-host parity before release. Trigger on "multi-platform plugin", "portable plugin", "make this plugin work in Claude/Codex/Hermes/OpenCode/Antigravity", "cross-agent compatibility", "plugin parity", "gap analysis", "compatibility matrix", or "drift" when more than one agent runtime is involved.
Use this skill when creating, updating, debugging, or publishing Google Antigravity plugins for the Antigravity App, Antigravity CLI (agy), and Antigravity IDE — scaffolding plugin.json, packaging skills/, agents/, rules/, mcp_config.json, hooks.json, and sidecars/, wiring hook payloads (PreToolUse decisions, injectSteps, Stop continuation), configuring Streamable HTTP or stdio MCP servers with Google ADC or OAuth, tuning rule triggers (always_on, model_decision, glob, manual), and installing locally, from GitHub, or from antigravity-plugins-official. Use when a plugin must load in .agents/plugins/ or ~/.gemini/config/plugins/, when agy plugin list shows missing components, when a rule is silently ignored, or when converting Codex/Claude packaging to Antigravity.
Use this skill when creating, updating, debugging, validating, or publishing Claude Code plugins and mods — scaffolding .claude-plugin/plugin.json, authoring skills, commands, subagents, hooks, MCP/LSP servers, monitors, themes, output styles, and bin/ executables, writing a mods hooks module (register, ui.render, $ API), declaring dependencies and release tags, running claude plugin validate, claude plugin test, claude plugin eval, and --plugin-dir sessions, or migrating a standalone .claude/ setup into a distributable plugin. Trigger on "Claude Code plugin", "plugin manifest", "plugin.json", "mods", "hooks module", "plugin marketplace", "plugin eval", "plugin validate", or packaging workflows for distribution.
Use this skill when the user wants to create, scaffold, update, migrate, configure, test, validate, or publish plugins for OpenAI ChatGPT and Codex — covering the skills/MCP/UI shape decision, plugin.json manifests, mcp.json server wiring, MCP tool contracts and safety annotations, OAuth 2.1 authentication, MCP Apps UI and ChatGPT extensions, MCP Events webhooks, lifecycle hooks, agents/openai.yaml skill dependencies, local and repo marketplaces, .codex/config.toml activation, and validation via bundled scripts. Also use when converting legacy .codex-plugin/ projects, connecting an MCP server to a plugin, or preparing a workspace or public-directory submission.
Create, update, scaffold, or validate plugins for Hermes Agent (NousResearch/hermes-agent) — general plugins (tools, lifecycle hooks, slash commands, hermes <plugin> CLI subcommands, bundled skills, message injection) and specialized plugins (model provider, memory provider, context engine, gateway platform, image-gen, video-gen). Covers plugin.yaml manifests, the register(ctx) wiring contract, handler rules, discovery/opt-in semantics, and validation with hermes plugins, HERMES_PLUGINS_DEBUG=1, and hermes doctor. Use when building a Hermes plugin end-to-end, adding a tool or hook, injecting turn context, adding slash/CLI commands, building a provider backend, or debugging why a plugin is not discovered or not enabled.
Use this skill when creating, updating, debugging, or configuring OpenCode V2 plugins: local plugin directories under .opencode/plugins/ or ~/.config/opencode/plugins/, published packages in opencode.json(c), server plugins (Plugin.define setup/effect), CLI/TUI extensions, and embedded @opencode/client or @opencode/sdk integrations. Covers plugin discovery and enablement, synchronous domain transforms (agent, provider, model, tool, command, skill, MCP, VCS, worktree, ...), runtime hooks (session, tool, shell, permission), custom tools, storage, RPC contracts, Cloudflare workerd, and OpenTUI 0.6. Use when customizing domains at runtime, protecting files or shell commands, publishing packages with ./rpc or ./tui exports, or debugging plugin load/execution.