Source of truth: and Last updated: 2026-05-13 — v3.8.0

---

Built-in skills () — shipped with OmniRoute. Cover the common cases:
, — per-API-key sandbox workspace under
• — outbound HTTP through with
• — pluggable search provider with caching ()
• — Docker-sandboxed or execution
• — Docker-sandboxed shell command
• — Playwright-backed scaffolding, disabled by default ()
1. SkillsMP (the OmniRoute Marketplace) — fetched from . Requires in Settings.
2. SkillsSH ( community catalog) — fetched from . No auth needed; SKILL.md content pulled from GitHub raw.

). Switch it under Settings → Memory & Skills. Default: .

in the in-memory registry (). Version must be semver (). understands , , , , , , , and exact-match constraints.

	(default 3)

is the default for marketplace-installed skills. and together mean "registered but inactive" — toggling via the legacy column also bumps so older codepaths stay consistent ().

table with the following statuses ():

enum SkillStatus {
  PENDING = "pending",
  RUNNING = "running",
  SUCCESS = "success",
  ERROR = "error",
  TIMEOUT = "timeout",
}

is a singleton with a 60-second TTL cache (). is idempotent and dedupes concurrent calls via . Any write (//) invalidates the cache. Look up versions via and .

in is the entry point that turns registered skills into provider-specific tool definitions:

OpenAI —
• Anthropic —
• Google (Gemini) —

so the handler can pick the right version when the model calls it back.

, each candidate skill is scored against the request context ( in ):

skills with are injected. Ties are broken by (desc), then alphabetical name ().

in is invoked by the chat handler after the upstream returns a tool-calling response:

reads provider-specific shapes (OpenAI / Responses , Anthropic , Gemini ).
1. → ) are resolved first. Built-in handlers run inline.
2. .
3. , items, or Anthropic blocks as appropriate.

in the execution context can be set to to allow only built-in interception (used by request paths that explicitly disable user-defined handlers).

---

, ) run inside Docker via (). Every container is launched with:

):

		+
		and are tmpfs

and are exposed for shutdown; running containers are tracked in .

in :

	(1 MB)	and
		response body
		or to allow per-call opt-in

, , . Any additions via are merged with the defaults; unknown images are rejected by .

env var. The non-sandbox handler timeout is hard-coded to 30 s in () but can be overridden at runtime via .

and resolve every path relative to a per-API-key workspace at . Path traversal () and forbidden segments (, , , , , ) are rejected before any disk I/O.

():

• )
• with (private/loopback ranges blocked)
• ; client sees

---

defines a that decides between (in-process) and execution per call, with an retry path on timeout/memory errors. The wired-in / implementations are stubs (, return placeholder objects) — treat this module as a contract under construction. Real execution still goes through + .

---

— base and tables, with indexes on and .
• — adds , , (JSON), to .

is constrained at the database level: .

---

. Management endpoints (, , ) require management auth via . The marketplace/install flows use the lighter (session or API key).

| GET | List registered skills. Supports , , , pagination | | | PUT | Update or | | | DELETE | Unregister by id | | | POST | Install a custom skill (handler code + schema) | | | GET | Search the SkillsMP catalog (returns popular defaults when is empty) | | | POST | Install a SkillsMP skill (requires active provider = ) | | | GET | Search the skills.sh catalog (, capped at 100) | | | POST | Install a skills.sh skill (requires active provider = ) | | | GET | Paginated execution history () | | | POST | Execute a registered skill ad-hoc |

endpoint returns HTTP with when (). Operators can flip the master switch from Settings → AI.

curl -X POST http://localhost:20128/api/skills/install \
  -H "Authorization: Bearer $OMNIROUTE_MGMT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "reverse-text",
    "version": "1.0.0",
    "description": "Reverses a string",
    "schema": {
      "input":  { "type": "object", "properties": { "text": { "type": "string" } }, "required": ["text"] },
      "output": { "type": "object", "properties": { "reversed": { "type": "string" } } }
    },
    "handlerCode": "echo-handler",
    "apiKeyId": "your-api-key-id"
  }'

string is a handler name lookup — not executable code. The executor maps it via (). Marketplace installs store the SKILL.md text in this field as documentation and route execution through model-generated tool calls. Arbitrary user-supplied source is not eval'd.

---

). They are auto-registered when the MCP server boots.

	, ,

MCP-SERVER.md for transport setup and scope assignments.

---

exports the A2A skill descriptor and a helper. Custom A2A skills live in and are dispatched via (). See A2A-SERVER.md for the full task lifecycle.

---

Define the handler in (or a sibling file under ). Signature: .
1. Sandboxed code path? Call . Use against the allowlist.
2. Filesystem path? Always pass through before touching disk.
3. Network call? Use with ; sanitize headers via .
4. Register by adding the entry to (or calling -style at boot).
5. Wire built-in tool aliases (optional) in () if the upstream model emits a different name.
6. Tests in (Vitest).

---

skillExecutor.registerHandler("my-handler", async (input, ctx) => { ... });
1. (the field must match the registered handler name).
2. to or via .

---

Master switch: blocks all execution and returns HTTP on . The registry continues to load.
• Lock down egress: keep unset (default) for fully air-gapped sandboxing. Per-call still requires the master gate.
• Allow specific images: set to extend the allowlist.
• Audit executions: and both query . Successful runs include ; failures include .
• Cache invalidation: call after manual DB edits; otherwise wait 60 s.
• Anonymous workspace: when is empty, all calls hash to the same workspace — share-aware code should always pass a real key.

---

MCP-SERVER.md — MCP tool registration and transports
• A2A-SERVER.md — A2A task lifecycle and skill dispatch
• USER_GUIDE.md — user-facing introduction
• ARCHITECTURE.md — request pipeline and component map
• , ,