Skill for developing AI agent engine components in internal/agent/. Covers the ReAct loop, swarm delegation, planning, verification, memory reflection, and cron automation.
The English summary is being prepared. Showing the original description.
Skill for developing REST API endpoints in internal/server/. Covers routing, request validation, WebSocket streaming, and API conventions.
The English summary is being prepared. Showing the original description.
See how to installReview the instructions the agent will receive before you install.
Use this skill when creating or modifying REST API endpoints in the internal/server/ package.
internal/server/
├── router.go # Chi router setup, global & auth middlewares, route tree
├── api_auth.go # Setup, login, logout, password change, auth status
├── api_dashboard.go # Dashboard aggregate metrics & summaries
├── api_agent.go # Agent CRUD, start/stop, chat, soul, memory-md, memories (pin/importance), cron
├── api_tasks.go # Autonomous Task matrix CRUD, Heartbeat config & manual pulse triggers
├── api_conversations.go # Chat conversations and message history
├── api_plugins.go # WASM plugin upload, enable/disable, logs, configuration & vault secrets
├── api_vault.go # Hardware-bound vault secret management
├── api_integrations.go # Channel accounts, pairing codes, sender authorization
├── api_tools.go # MCP servers, skills, tool execution, hub marketplace
├── api_workspace.go # Workspace file browser, read/write/mkdir/upload
├── api_system.go # Metrics, LLM health & retune, token usage ledger history, keys, identity, HAL
├── api_setup.go # Legacy/standalone setup endpoints
├── layered_fs.go # Layered filesystem (/data/overrides/ → go:embed fallback)
├── static.go # Embedded static asset server
└── server_test.go # Comprehensive endpoint test suite
/api (all routes are prefixed with /api)/api/v1/ prefix until v1.0.0 is officially released.ActonOS uses Chi v5 with standard library net/http:
import (
"github.com/go-chi/chi/v5"
"github.com/go-chi/chi/v5/middleware"
)
Success (s.respondJSON(w, http.StatusOK, data)):
{
"data": { ... }
}
Error (s.respondError(w, http.StatusBadRequest, "INVALID_REQUEST", "description")):
{
"error": {
"code": "INVALID_REQUEST",
"message": "Human-readable explanation of error"
}
}
Server// 1. Respond with JSON data wrapped in {"data": ...}
func (s *Server) respondJSON(w http.ResponseWriter, status int, data any)
// 2. Respond with standard error envelope wrapped in {"error": {"code": ..., "message": ...}}
func (s *Server) respondError(w http.ResponseWriter, status int, code, message string)
// 3. Decode request body with 1MB safety limit
func (s *Server) decodeJSON(r *http.Request, v any) error
GET /api/healthGET /api/modelsGET /api/notifications/push/vapid-keyGET /api/auth/statusPOST /api/auth/setupPOST /api/auth/loginPOST /api/auth/logoutAll other routes are nested inside r.Group with r.Use(s.RequireAuthMiddleware). Requests must include Authorization: Bearer <token> in the HTTP headers when authentication is initialized.
type UpdateSoulRequest struct {
SoulContent string `json:"soul_content"`
}
type SoulResponse struct {
AgentID string `json:"agent_id"`
Content string `json:"content"`
UpdatedAt string `json:"updated_at"`
}
*Serverfunc (s *Server) handleSaveSoul(w http.ResponseWriter, r *http.Request) {
agentID := chi.URLParam(r, "agentID")
if agentID == "" {
agentID = agent.DefaultSystemAgentID
}
var req UpdateSoulRequest
if err := s.decodeJSON(r, &req); err != nil {
s.respondError(w, http.StatusBadRequest, "INVALID_BODY", "failed to decode json body")
return
}
if err := s.profileMgr.SaveSoul(r.Context(), agentID, req.SoulContent); err != nil {
s.respondError(w, http.StatusInternalServerError, "SAVE_FAILED", err.Error())
return
}
s.respondJSON(w, http.StatusOK, SoulResponse{
AgentID: agentID,
Content: req.SoulContent,
UpdatedAt: time.Now().UTC().Format(time.RFC3339),
})
}
internal/server/router.gor.Route("/agents", func(r chi.Router) {
// ...
r.Route("/{agentID}", func(r chi.Router) {
// ...
r.Put("/soul", s.handleSaveSoul)
})
})
docs/API.md.agents/rules/source-registry.mdweb/src/lib/api.ts and interfaces to web/src/lib/types.tsFor streaming LLM tokens, reasoning thoughts, and tool call progress:
func (s *Server) handleChatStream(w http.ResponseWriter, r *http.Request) {
flusher, ok := w.(http.Flusher)
if !ok {
s.respondError(w, http.StatusInternalServerError, "STREAMING_UNSUPPORTED", "streaming not supported")
return
}
w.Header().Set("Content-Type", "text/event-stream")
w.Header().Set("Cache-Control", "no-cache")
w.Header().Set("Connection", "keep-alive")
eventChan := make(chan agent.AgentStreamEvent, 64)
go func() {
_, _ = s.engine.ExecuteStepStreamWithHistory(
r.Context(), agentID, msg, history, eventChan,
)
}()
for ev := range eventChan {
data, _ := json.Marshal(ev)
fmt.Fprintf(w, "event: %s\ndata: %s\n\n", ev.Type, data)
flusher.Flush()
}
}
The stream endpoint must flush live thought, token, tool_call,
tool_result, audit, done, and error events. It must not proxy to the
non-streaming JSON handler. Conversation messages are persisted before and
after the stream.
GET /api/realtime is a protected, same-origin WebSocket.actonos_token cookie set by setup/login; never put bearer tokens in WebSocket query strings.router.gos.respondError with appropriate HTTP statusweb/src/lib/types.ts and api.ts are synceddocs/API.md and .agents/rules/source-registry.md are updatedgo test ./internal/server/... passesapi_approvals.go owns durable exact-action approval decisions.api_runs.go exposes durable run summaries and ordered execution events.ToolRegistry.Execute and return HTTP 202 for ApprovalRequiredError.requestAdminApproval and exact dispatch from api_approvals.go.GET /api/system/audit/verify is the canonical audit-chain integrity check.Config.DataDir,
Config.WorkspaceDir, Config.SkillsDir, or Config.WASMDir; handlers must
not assume the process working directory is the data root.Server.vault; never write API keys into JSON
or .key files. Preserve automatic legacy migration and fail closed without
Vault.VACUUM INTO through the live database connection;
never copy the main database file while WAL mode is active.No reviews yet. Tell others what you thought.
No login needed. Share how it worked for you, ask a question, or note where it did not run.
No comments yet. Be the first to write one.
Skill for developing AI agent engine components in internal/agent/. Covers the ReAct loop, swarm delegation, planning, verification, memory reflection, and cron automation.
The English summary is being prepared. Showing the original description.
Skill for building ActonOS: Web UI → Go binary → Docker image → ISO. Covers the entire build pipeline.
The English summary is being prepared. Showing the original description.
Skill for developing messaging channel adapters via WASM Plugins and managing routing, session state, pairing codes, and accounts in internal/channels/.
The English summary is being prepared. Showing the original description.
Skill for developing the ActonOS React 19 + Tailwind v4 frontend adhering to the official DESIGN.md specification. Covers design tokens, component decomposition, zero-hardcoded-text i18n, and go:embed delivery.
The English summary is being prepared. Showing the original description.
Skill for creating ActonOS releases. Covers version bumping, changelog management, git tagging, Docker image publishing, and ISO creation.
The English summary is being prepared. Showing the original description.
Skill for initial ActonOS project setup, environment verification, and dependency installation.
The English summary is being prepared. Showing the original description.