ikanos-capability

name: ikanos-capability version: "1.0.0-beta1" description: > Skill for authoring, validating, and debugging Ikanos Capability YAML files (spec v1.0.0-beta1). Activate when the user wants to: write a new capability document, add or change authentication on a consumed API, configure orchestration steps or parameter mappings, set up a forward proxy, expose an MCP server or Skill server, configure external references for secrets, add a control port for health checks and metrics, enable OpenTelemetry observability, or run the Spectral linter. The Ikanos Specification defines modular, composable capabilities that consume external APIs and expose REST, MCP, Skill, or Control adapters. allowed-tools: - Read - Write - Bash - Glob

Overview

Ikanos lets you declare capabilities — functional units that consume external APIs and expose adapters (REST, MCP, Skill). A capability is a single YAML file validated against the Ikanos JSON Schema (v1.0.0-beta1).

Key spec objects you will work with:

• Info — metadata: display, description, tags, stakeholders
• Capability — root technical config; contains exposes, consumes, and aggregates
• Consumes — HTTP client adapter: baseUri, namespace, resources, operations
• Exposes — server adapter: REST (type: rest), MCP (type: mcp), Skill (type: skill), or Control (type: control)
• Aggregates — DDD-inspired domain building blocks; each aggregate groups reusable flows under a namespace. Tools and operations reference flows via ref
• Observability — optional OTel configuration on the control adapter: trace sampling, propagation format, OTLP exporter endpoint
• Script Steps — embed JavaScript, Python, or Groovy transformations between API calls using sandboxed GraalVM engines. Scripts read step results via bound variables and produce output through a result variable
• Binds — variable injection from file (dev) or runtime (prod)
• Namespace — unique identifier linking exposes to consumes via routing

Canonical sources (read these, never duplicate them):

• Specification: ikanos-docs/wiki/Specification.md
• JSON Schema: ikanos-spec/src/main/resources/schemas/ikanos-schema.json
• Polychro Ruleset: ikanos-spec/src/main/resources/rules/ikanos-rules.yml

Named-Entity Format (beta1)

Since v1.0.0-beta1, most named collections use keyed maps instead of arrays. The entity's name becomes the YAML map key; the name property is removed from the object body.

Keyed maps (key = name, no name property inside): tools, resources, operations, inputParameters (all contexts), steps, flows, aggregates, skills, prompts

Arrays (keep - item syntax): outputParameters, binds, exposes, consumes, stakeholders, mappings

Example — MCP tool (keyed map for tool + keyed map for inputParameters + array for outputParameters):

tools:
  list-ships:                          # ← key IS the name
    description: "List ships"
    inputParameters:                   # ← keyed map (name is the key)
      status:
        type: string
        description: "Filter by status"
    call: registry.list-ships
    outputParameters:                  # ← always array
      - type: array
        mapping: "$."

Example — REST resource + operation (keyed maps for both + keyed map for inputParameters):

resources:
  ships:                               # ← key IS the name
    path: "/ships"
    operations:
      list-ships:                      # ← key IS the name
        method: GET
        inputParameters:
          status:                      # ← key IS the name (REST/consumes)
            in: query
            type: string
        outputParameters:              # ← always array
          - type: array
            mapping: "$."

Example — aggregates + flows (keyed maps):

aggregates:
  forecast:                            # ← key IS the aggregate namespace
    description: "Weather forecast domain"
    flows:
      get-forecast:                    # ← key IS the flow name
        description: "Retrieve forecast for a location"
        inputParameters:
          location:                    # ← key IS the name
            type: string
            required: true
        call: weather-api.get-forecast

Decision Framework

Match the user's situation to a story reference. Each story explains why (the user's problem), what (the Ikanos pattern), and points to the spec for how.

If the user's intent does not match any story, read the canonical Specification directly.

Workflows

Edit a Capability

1. Analyze and propose. Infer the story from context (user prompt, open files, existing capabilities). Read that story file, then present a capability outline for the user to validate. Only ask what you cannot infer.
2. Scaffold. Copy assets/capability-example.yml. The document must begin with ikanos: "1.0.0-beta1".
3. Fill exposes. Choose the adapter type (REST, MCP, or Skill) and follow the pattern from the story. For REST operations, use call + with (simple) or steps + mappings (orchestrated) — never both.
4. Fill consumes. For each external API: set type: "http", a unique namespace (kebab-case), baseUri (no trailing slash), resources, and operations. Add authentication if needed.
5. Add binds if secrets or environment variables are needed. Use location: "file:" for dev; remove it for prod.
6. Review. Read references/design-guidelines.md and check your document against the design guidelines.
7. Validate. Run bash scripts/lint-capability.sh path/to/capability.yml. Fix any errors, then re-lint until clean.

Lint a Capability

1. Run the lint script: bash scripts/lint-capability.sh path/to/capability.yml Do NOT regenerate or modify this script.
2. Spectral reports errors and warnings with rule names. Common rules:
   • ikanos-namespaces-unique (error) — duplicate namespace
   • ikanos-consumes-baseuri-no-trailing-slash (warn) — trailing /
   • ikanos-consumed-resource-no-query-in-path (warn) — query in path
   • ikanos-rest-resource-path-no-trailing-slash (warn)
   • ikanos-baseuri-not-example (warn) — placeholder URI
   • ikanos-no-script-tags-in-markdown (error) — XSS in descriptions
   • ikanos-consumes-description (warn) — missing description
   • ikanos-control-port-singleton-and-unique (error) — more than one control adapter, or port collision
   • ikanos-control-address-localhost-warning (warn) — control port bound to non-localhost For the full rule list, read the Spectral ruleset file directly.
3. Fix and re-lint. Repeat until clean.

Mock a Capability

Use mock mode when the upstream API does not exist yet, or you want to prototype a contract-first design. Read references/mock-capability.md before writing any mock output parameters.

Key rules:

1. Omit call, steps, and the entire consumes block — they are not needed for a pure mock capability.
2. Use value on MappedOutputParameter for static strings (value: "Hello") or Mustache templates (value: "Hello, {{name}}!"). Do NOT use const — it is schema-only and is never used at runtime.
3. Mustache placeholders in value are resolved against the tool's or endpoint's input parameters by name. Only top-level input parameter names are in scope — no nesting, no with remapping.
4. value and mapping are mutually exclusive on a scalar output parameter — never set both.
5. Object and array type output parameters in mock mode must carry value on each leaf scalar descendant; the container itself has no value.
6. When the mock is ready to be wired to a real API, add consumes, replace value with mapping, and add call or steps — the exposed contract does not change. and trustedHeaders (at least one entry).
7. MCP tools must have name and description (unless using ref, in which case they are inherited from the referenced aggregate flow). MCP tool input parameters must have name, type, and description. Tools may declare optional hints (readOnly, destructive, idempotent, openWorld) — these map to MCP ToolAnnotations on the wire.
8. ExposedOperation supports three modes (oneOf): simple (call + optional with), orchestrated (steps + optional mappings), or ref (ref to an aggregate flow). Never mix fields from incompatible modes.
9. Do not modify scripts/lint-capability.sh unless explicitly asked — it wraps Spectral with the correct ruleset and flags.
10. Do not add properties that are not in the JSON Schema — the schema uses additionalProperties: false on most objects.
11. Every exposed inputParameter must be referenced in at least one step's with block or mapping — orphan parameters bloat the API surface and confuse consumers.
12. Every consumed outputParameter must be referenced in the exposed part (via mappings, or outputParameters) — unreferenced outputs are dead declarations that add noise without value.
13. Do not prefix variable names with the capability, namespace, or resource name — variables are already scoped to their context. Redundant prefixes reduce readability without adding disambiguation.
14. When using ref on MCP tools or REST operations, the ref value must follow the format {aggregate-namespace}.{flow-name} and resolve to an existing flow in the capability's aggregates section.
15. Do not chain ref through multiple levels of aggregates — ref resolves to a flow in a single aggregate, not transitively.
16. Aggregate flows can declare semantics (safe, idempotent, cacheable). When exposed via MCP, the engine auto-derives hints from semantics. Explicit hints on the MCP tool override derived values.
17. Do not duplicate a full flow definition inline on both MCP tools and REST operations — use aggregates + ref instead.
18. In mock mode, use value on output parameters — never const. const is a JSON Schema keyword retained for validation and linting only; it has no effect at runtime.
19. In mock mode, Mustache templates in value fields resolve only against top-level input parameter names. Do not reference with-remapped consumed parameter names — those are not in scope for output resolution.
20. At most one type: "control" adapter is allowed per capability. Its port must not collide with any business adapter port.
21. The control port address should be localhost or 127.0.0.1 for security. Binding to 0.0.0.0 exposes management endpoints externally.
22. The /metrics and /traces control port endpoints are configured under observability.metrics.local and observability.traces.local respectively. Observability is enabled by default — all enabled fields default to true. Set enabled: false to disable specific endpoints or observability entirely. They return 503 when the OTel SDK is not on the classpath.
23. When adding a control port, also add observability on the control adapter if you want metrics and traces to produce data. Without it, the local endpoints return empty or 503 responses.
24. The observability.exporters.otlp.endpoint field supports Mustache expressions for binds (e.g. "{{OTEL_ENDPOINT}}"). Use binds to keep exporter URLs environment-specific.
25. Script steps require a file property. language and location are optional when Control Port defaults are configured via management.scripting.defaultLanguage and management.scripting.defaultLocation.
26. Script dependencies are pre-evaluated in order before the main script. Shared logic (helpers, constants) goes in dependencies.
27. The script must assign to the result variable to produce output. Previous step results are bound as variables matching step names.
28. When management.scripting is configured on the Control Port, defaultLanguage and defaultLocation serve as fallbacks — step-level values take precedence.
29. allowedLanguages restricts which languages script steps may use. If omitted, all three languages (javascript, python, groovy) are allowed.