inference-gateway

Drive interactive terminal programs (TUIs, REPLs, pagers, a debugger, another CLI's chat UI) that the Bash tool cannot script directly, by running them inside tmux and scripting them with send-keys / capture-pane. Use whenever a task needs you to launch, type into, read the live screen of, or tear down an interactive terminal program - e2e-testing a TUI, stepping a debugger, or answering a prompt a plain command would block on. Not for subagents you spawned with the Agent tool (use ReadSubagentScreen / SendSubagentInput for those).

Build interactive terminal forms and prompts with huh v2 (charm.land/huh/v2): Form/Group/Field composition, Select/MultiSelect/Input/Confirm/Text/FilePicker, embedding forms inside a Bubble Tea v2 model, standalone Run() prompts, theming, and filtering. Use when adding or reviewing any interactive form, picker, wizard, confirm, or approval prompt with this stack. v2's Theme became an interface and forms complete via internal command messages, so lead with the contract and the v1->v2 table - never write from v1 memory.

Build, consume, debug, or reason about Agent2Agent (A2A) protocol agents - Agent Cards and discovery, the Task lifecycle, Messages/Parts/Artifacts, the three transport bindings (JSON-RPC 2.0, gRPC, HTTP+JSON), request/response vs streaming (SSE) vs push notifications, security schemes, webhook security, and A2A error codes. Use whenever the work touches A2A: writing or reviewing an agent server or client, designing an agent card, delegating tasks between agents, wiring the CLI's A2A tools, or answering "how does A2A do X". The source of truth is the schema in inference-gateway/schemas - reference it, never hand-copy it, and always pin the spec version you target (v0.3.0 vs v1.0 differ in breaking ways).

Build and review Go CLI apps with Cobra and Viper (both used heavily here) - command-first architecture, RunE error handling, PersistentPreRunE setup, flag design, typed-struct config, the flag/env/file/default precedence hierarchy, and in-memory command tests. Use when adding commands or flags, wiring Viper config or env binding, or reviewing any cmd/ code. Distilled from spf13/go-skills (by Cobra's and Viper's author) and adapted to this repo's config layering.

Use the Bubbles v2 component library for Bubble Tea v2 (charm.land/bubbles/v2): viewport, textinput/textarea, spinner, progress, list, table, key, help. Use when adding or reviewing prebuilt TUI components - scrollable viewports, text inputs, spinners, progress bars, key bindings and help - with this stack. Each component is a sub-model you embed and delegate to; v2 constructors and key handling differ from v1, so lead with the contract and the v1->v2 table.

Build and review Bubble Tea v2 terminal UIs in Go (charm.land/bubbletea/v2) - The Elm Architecture: a tea.Model with Init/Update/View plus tea.Cmd/tea.Msg. Use whenever writing or reviewing a TUI, a tea.Model, Update/View methods, commands, key/mouse handling, or full-screen terminal apps with this stack, even if "Bubble Tea" isn't named. v2's API differs sharply from v1 - lead with the v1->v2 table so you don't write v1 code that won't compile.

Style and lay out terminal output in Go with Lip Gloss v2 (charm.land/lipgloss/v2) - styles, colors, borders, layout joins, and tables. Use when writing or reviewing terminal styling: foreground/background colors, padding/border/width, JoinHorizontal/ JoinVertical, Place, or lipgloss/v2/table, usually alongside Bubble Tea v2. v2's color and renderer model changed sharply from v1 - lead with the v1->v2 table so you don't reach for AdaptiveColor or a global renderer that no longer exist.

Idiomatic Go concurrency - sync primitives, channel semantics, the select statement, and the standard channel patterns (cancellation/done-channel, fan-out/fan-in, pipeline, or-done, context, worker-pool). Use when writing or reviewing Go that spawns goroutines, shares state, or coordinates over channels, even if the user never says "concurrency" - e.g. data races, deadlocks, goroutine leaks, sync/context/select usage, or "make this parallel". Distilled from luk4z7/go-concurrency-guide.

Idiomatic Go - package and interface design, error wrapping, table-driven tests, generics, the modern standard library (slices/maps/cmp/errors.Join), current syntax, and logging discipline. Use when writing, reviewing, or refactoring any Go code, especially code drifting toward Java/Spring shapes (deep layer trees, generic repositories, heavy frameworks) - even if the user never says "idiomatic". Distilled from spf13/go-skills and adapted to this repo.

Review a Go design spec before implementation begins - dispatch a subagent that checks a design doc for completeness, consistency, and idiomatic Go (simplicity, small consumer-defined interfaces, explicit wrapped errors, context propagation) plus Cobra/Viper conventions where applicable. Use when a user has a Go spec to review, asks "is this spec ready?", or is about to implement from a written design. Distilled from spf13/go-skills and adapted to this repo's conventions.

Use this when the user wants to verify the agent is reachable and responding correctly. Invokes the echo tool with a known payload and confirms the round-trip succeeded.

Use this when the user wants to test how their client handles different failure modes. Invokes the error tool across the supported error_type values (validation, timeout, internal, not_found) so the caller can observe each error path.

Use this when the user wants to test client behavior under slow responses with realistic payloads. Combines the delay tool (to introduce latency) with the random_data tool (to produce a test payload of the requested shape).

Author and maintain projects built with ADL (Agent Definition Language) - the YAML manifest format from `inference-gateway/adl` that `adl-cli` turns into a full A2A agent scaffold (Go/Rust/TypeScript). Use when working inside a generated agent project (presence of `agent.yaml` + `.adl-ignore`), when editing the manifest, when implementing custom tools and skills against the generated scaffold, or when planning the schema-first / domain modelling for a new agent.

Author a new Agent Skill from a minimal SKILL.md template. Use when the user wants to create a new skill, scaffold one from scratch, or learn the SKILL.md frontmatter and body conventions.

Use this when the user asks to schedule a meeting, book a slot, or "find a time that works". Resolves a conflict-free booking by finding open slots, validating no overlap, and creating the event.

Use this when the user requests a new n8n workflow or asks to automate a process. Searches relevant nodes with search_n8n_docs, drafts the workflow YAML, validates it with validate_n8n_workflow, then saves it via create_artifact.

Use this when you need up-to-date documentation for a third-party library or framework before writing code against it. First resolves the library name to a Context7-compatible ID via resolve_library_id when the caller does not already know it (format '/org/project' or '/org/project/version'), then fetches focused, topic-scoped documentation via get_library_docs. Good for filling in unknowns about specific APIs, hooks, configuration options, or version-specific behavior.