[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M735HSSTZN76TXAR7PE5YC71":3,"skill-same:01M735HSSTZN76TXAR7PE5YC71":51},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":29,"commitSha":30,"skillDirectory":31,"forksCount":14,"repositoryHtmlUrl":32,"repositoryHomepageUrl":6,"ownerHtmlUrl":33,"ownerBlogUrl":34,"ownerDescription":35,"updatedAt":36,"install":37,"resources":50},"01M735HSSTZN76TXAR7PE5YC71","architecture",null,"Reuse org catalog resources when deriving or enriching a component's design — deciding the component decomposition, filling a scaffolded design.json (language, dependencies, description, pinned skills), or resolving\u002Freconsidering any dependency.","wso2\u002Flabs-agentic-engineer","wso2","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F533043?v=4","01M6CDAFS40T9Y86DYXWE0NJES","WSO2","01M735EE4N2PP2ZR56AX7DNPNP",24,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Fwso2\u002Flabs-agentic-engineer\u002Fblob\u002Fmain\u002Fskills\u002Farchitecture\u002FSKILL.md",[],[],"2026-11-11 15:11:14",false,"---\nname: architecture\ndescription: \"Reuse org catalog resources when deriving or enriching a component's design — deciding the component decomposition, filling a scaffolded design.json (language, dependencies, description, pinned skills), or resolving\u002Freconsidering any dependency.\"\nmetadata:\n  aep:\n    kind: platform\n    audience: [design]\n---\n\n# Architecture\n\nComponent decomposition and per-component facts. The cell (`cell-design`\nskill) declares WHAT exists; this skill owns deciding the decomposition and\nENRICHING each component's `design.json`.\n\n## Scaffold first, enrich second\n\nWhen the cell is saved, the platform scaffolds\n`specs\u002Fdesign\u002Fcomponents\u002F\u003Cid>\u002Fdesign.json` for every deployable component —\nthe mechanical fields (name, type, version, buildpack, appPath, entrypoint, a\ndefault exposure) plus a `\"language\": \"TBD\"` sentinel. Your job is the\nJUDGMENT fields, and the build gate refuses a deployable component left\nunenriched:\n\n- **language** — **Language is decided in this order**: the organization skill's Tech stack\ndefault first, then a language the requirements name, then the platform\ndefault — **Ballerina** for backend services, TypeScript for web apps.\nReplace the scaffold's `\"language\": \"TBD\"` sentinel with the decided value;\nthe build gate refuses a component whose language is still TBD. That choice is a fact\nyou record, not a preference you re-derive per component: write it as the\ncomponent's `language` and pin the matching stack skill in `skillsPinned`.\n- **stories** — the IDs of the stories this component serves, as written in\n  the feature files (e.g. `\"stories\": [\"F1.1\", \"F1.2\", \"F2.1\"]`). Claim every\n  story the component actually serves: the build gate refuses the tag while\n  any story is claimed by no component, or an ID the requirements do not\n  define is cited.\n- **dependencies** — the playbook below.\n- **description** — one paragraph: single responsibility, port\u002Fentrypoint\n  expectations, and what it explicitly does NOT do.\n- **skillsPinned** — After emitting or changing a component's design, record the skills that\ncomponent's build actually needs as a `skillsPinned` array **inside that\ncomponent's `specs\u002Fdesign\u002Fcomponents\u002F\u003Cname>\u002Fdesign.json`** — use the exact\ncatalog names, e.g. a Ballerina API service →\n`[\"openapi-conventions\", \"ballerina\"]` (a Go one → `[\"openapi-conventions\",\n\"go\"]`); a web-application → `[\"react-webapp\", \u003Cdesign system>]`,\nwhere `\u003Cdesign system>` is the skill name in the **UI design system** section of\nthe Organization defaults block in your instructions. A web app's UI toolkit is\na settled organization decision, so pin it on **every** web-application rather\nthan leaving it to a description to trigger — and read the name from that\nsection every time instead of remembering one, because it differs per\norganization. If that section is absent or empty, pin only `react-webapp`.\nNever substitute a design system the organization defaults do not name.\n\n**Pin the design system; do not consult it.** A design system is built against,\nnot designed with — it is the coding run's to load, and its theming is a settled\norganization decision that no design-time question reopens. Never ask the user\nabout colors, themes, or look and feel. An ai-agent → `[\"agent-building\"]`. **Pin the auth skills the same way,\nnever leave them to a description:** `\"thunder-authentication\"` on **both**\nsides of sign-in — every component that declares the `thunder-app` dependency,\nthe SPA *and* each protected backend it calls — and `\"api-management\"` on every\nservice that sits behind the gateway. The first owns the SPA's sign-in, its\nscreen gates and the backend's verification of the gateway's assertion; the\nsecond owns the gateway's contract. It is a JSON key on the component's design\nobject, so include it when you write that `design.json` (addFile\u002FeditFile) —\n`design.json` is its only home.\nEach component carries only the skills its own build needs.\n\nWriting the whole enriched file yourself (removeFile + addFile with every\nfield) is equally valid — the scaffold is a safety net, not a required\nintermediate.\n\n## Deriving components — deployment units the requirements justify\n\nA component is one independently deployable unit, NOT a domain concept. The\nright number comes from the requirements: for every component you must be\nable to say \"this deploys and evolves independently because \u003Csomething the\nrequirements state>\". Write that justification into the component's\n`description`.\n\nA requirement justifies a SEPARATE component when it shows:\n\n- a distinct user-facing surface — e.g. an internal admin portal AND a\n  customer-facing app with different users and lifecycles → two web-applications;\n- a genuinely different runtime or scaling profile — e.g. an async\n  worker\u002Fbatch processor beside an interactive API, or a long-running\n  AI\u002Finference service;\n- a technology the rest of the system doesn't share — e.g. a Python ML\n  service beside a Ballerina API;\n- an explicitly separate lifecycle or ownership stated in the requirements.\n\nDo NOT split by:\n\n- entity or domain concept — claims-service, users-service,\n  receipts-service... is a domain model dressed as a topology; those are\n  modules of ONE service;\n- layer — auth, notifications, file storage as own services when they are\n  modules of the API; a single frontend split into `ui-shell` + `ui-pages` when\n  it is one bundle;\n- infrastructure — api-gateway, database, queue, auth server are NEVER\n  components; the platform provides them;\n- periodic work — a nightly digest or a cleanup sweep is a background task the\n  owning service starts, not a component of its own. Only a genuinely different\n  runtime or scaling profile (above) justifies splitting one off.\n\n**Never invent a user-facing surface the requirements did not ask for.** A\nsystem whose requirements describe a browser app lands at one service + one\nweb-application; a system whose only surface is an agent lands at one service +\none ai-agent and NO web-application. An agent is a complete surface on its own —\nthe platform's Try it app talks to a deployed agent directly — so \"there must be\nsomething for the user to open\" is not a reason to add a SPA. Adding an\nunrequested web-application is the same failure as splitting by domain concept:\na shape imposed on the requirements instead of read from them. Name components\nin kebab-case after their responsibility (`expense-api`, `expense-webapp`,\n`report-worker`, `packing-agent`).\n\n**An AI agent is `\"ai-agent\"`, and every model call is one.** It is the only\ncomponent type with model access. Reach for it when the requirements call for a\nconversational or autonomous surface — a user talking to the system in their own\nwords rather than filling in a form — and for every agent the requirements'\ndecisions name. Its behaviour is authored as\n`agent.afm.md` (the `agent-building` skill), it is implemented in TypeScript, and\nit pins `[\"agent-building\"]`. It is a normal deployable that calls other\ncomponents over HTTP: give it a `component` dependency for every API it uses.\n**An agent a signed-in user reaches is a protected backend** — whether the\ncaller is a sibling web-application or the platform's Try it app, set\n`\"exposure\": \"internet\"` (a browser cannot reach an intranet address, so an\nintranet agent has nothing that can talk to it) AND give it the project's\nshared `thunder-app`\ndependency, under the same dependency NAME the SPA and the sibling APIs use,\nexactly as you would for a service. It is the same sign-in and the same OAuth\napp, not a second one. Do this even when the agent stores nothing and every API\nbehind it already authorises: a service you can call without a token costs CPU,\nan agent you can call without a token costs money on the organisation's model\nkey, so an unauthenticated agent endpoint is a billing hole rather than a\ntolerable one.\n**Declare no dependency for model access** — every `ai-agent` gets it from its\ncomponent type, on the organisation's own key, so there is nothing to choose.\n**An agent whose work needs files declares them.** Give it `x-aep.attachments`\nwith the types and counts the feature needs (`agent-building` has the ceilings\nand the contract). One whose work runs with nobody signed in — a ticket from an\narriving email, a nightly job — does not fit yet: ask the user at design\naltitude whether a person does that step for now or the feature waits.\n**An `ai-agent` with server memory needs a Postgres for its conversation\nstore.** Give it a `platform-resource` dependency with `resourceType:\n\"postgres-cnpg\"` — the PVC-backed type, so a conversation survives a pod\nrestart; the older ephemeral `postgres` type is not installed and referencing\nit fails provisioning. Dedicated is the default — `{ \"kind\":\n\"platform-resource\", \"name\": \"memory-db\", \"resourceType\": \"postgres-cnpg\" }`,\ninjected as `MEMORY_DB_HOST` \u002F `MEMORY_DB_PORT` \u002F `MEMORY_DB_DBNAME` \u002F\n`MEMORY_DB_USER` \u002F `MEMORY_DB_PASSWORD` (`postgres-cnpg` has no `url` output;\nthe generic platform-resource wiring injects each of its five outputs as\n`\u003CDEP_NAME>_\u003COUTPUT>`, uppercased). When the project already carries a\nPostgres and the user prefers one instance, declare the SAME dependency name\nthe sibling service uses — same-name resolution to one shared instance is the\n`thunder-app` sharing rule, and it applies here too. Either way the agent owns\nits `conversations` table exclusively: no other component touches it, and the\nagent touches nothing else in a shared instance. Business data is still not\nthe agent's to hold — anything beyond its own conversation belongs behind a\nservice that owns it.\n\n**Component `type` is a fixed vocabulary — use the EXACT string.** A backend is\n`\"service\"`; a browser app is `\"web-application\"` (OpenChoreo's own term). Write\n`\"web-application\"` verbatim — NOT `\"webapp\"`, `\"web-app\"`, or `\"webApplication\"`\n(those are rejected, and a wrong value silently breaks the app's deployment and\nruntime config). The `-webapp` in a component NAME is fine; the `type` is still\n`\"web-application\"`. Other kinds the requirements imply (`\"scheduled-task\"`,\n`\"worker\"`, …) are captured verbatim, but the platform installs component types\nfor `service` and `web-application` ONLY — anything else records intent and does\nnot deploy, so reach for one when the requirements truly force it, never as the\ndefault home for periodic work.\n\n## Per-component design.json\n\nEach component's structured facts live in ONE JSON document (no markdown, no\nfrontmatter). The platform validates each write against this schema and rejects\nviolations:\n\n```json\n{\n  \"name\": \"expense-api\",              \u002F\u002F MUST equal the directory name\n  \"type\": \"service\",                  \u002F\u002F EXACT kind: \"service\" or \"web-application\" (NEVER \"webapp\"\u002F\"web-app\"), or another the requirements imply (\"scheduled-task\", \"worker\", ...)\n  \"version\": \"0.1.0\",                 \u002F\u002F semantic version; 0.1.0 for a new component\n  \"language\": \"Ballerina\",            \u002F\u002F implementation language — \"Ballerina\" for a service unless the requirements say otherwise; \"TypeScript\" for a web-application\n  \"buildpack\": \"docker\",              \u002F\u002F always \"docker\"\n  \"appPath\": \"expense-api\",           \u002F\u002F repo-relative source dir — the component name\n  \"entrypoint\": \"deployment\u002Fservice\", \u002F\u002F deploy entry — PAIRS with `type`: \"deployment\u002Fservice\" for a service, \"deployment\u002Fweb-application\" for a web-application, \"deployment\u002Fai-agent\" for an ai-agent\n  \"exposure\": \"internet\",             \u002F\u002F \"internet\" (public) | \"intranet\" (internal only)\n  \"dependencies\": [ \u002F* see below — every dependency edge touching this component appears here *\u002F ],\n  \"description\": \"One paragraph: single responsibility, port\u002Fentrypoint expectations, and what it explicitly does NOT do.\",\n  \"stories\": [\"F1.1\", \"F1.2\", \"F2.1\"], \u002F\u002F story IDs THIS component serves — the build gate refuses the tag while any story is claimed by nobody\n  \"skillsPinned\": [\"openapi-conventions\", \"ballerina\"], \u002F\u002F the skills this component's build needs — see the field above\n  \"endpoint\": { \"name\": \"http\" } \u002F\u002F optional; see below\n}\n```\n\n`name`, `type`, `version`, `language`, `buildpack`, `appPath`, `entrypoint`,\n`exposure`, `description`, `dependencies`, `stories` and `skillsPinned` are\nrequired — `stories` and `skillsPinned` are as required as the rest, and a\ncomponent missing either fails the build gate, so emit them in the SAME write\nrather than as a follow-up edit. To CHANGE a\ndesign.json, re-emit the whole corrected file (removeFile + addFile) — never\npatch JSON with anchored edits. On INVALID_JSON or SCHEMA_VIOLATION, fix what\nthe message lists and re-emit.\n\n`endpoint` is optional: omit it and a service's endpoint takes the default name\n`\"http\"`. Declare `{ \"name\": \"\u003Cendpoint-name>\" }` only when the endpoint must be\nnamed otherwise — `name` is the single source of truth the coding agent copies\ninto `workload.yaml` and the managed-API gateway binds to. The port lives in\n`workload.yaml`, not here.\n\n**Platform-owned fields you never author**, in two kinds:\n\n- **Preserved verbatim** where the platform has already written them:\n  `exposesAPI`, `componentAgentInstructions`, and any dependency\n  `status`\u002F`reason`.\n- **Recomputed and overwritten** on every save: a dependency's `wiring` object.\n  The platform derives its `ref` and its env-var names from the dependency's name\n  and its resource type's declared outputs, so anything you write there is\n  discarded. `stories` is NOT in this class — it is yours to author, per the\n  **stories** field above.\n\n### dependencies — one entry per design.cell edge\n\n`dependencies` mirrors the DEPENDENCY edges of design.cell: every arrow\nbetween this component and another node appears here and vice versa — a\nmismatch is a defect. The `north ->` \u002F `west ->` gateway arrows are\n**exposure**, not dependencies: they live in this component's `exposure`\nfield and never become entries here. The platform enforces the membership\nhalf as you write: a dependency whose `name` is not a node the cell declares\nis refused (`UNKNOWN_DEPENDENCY`) with the cell statement to add. A resource\nyou discover here — a database, a cache — is a cell node too: `editFile`\n`specs\u002Fdesign\u002Fdesign.cell` to add `component \u003Cname> as \"…\" database` inside\nthe cell before the design.json that depends on it, or the diagram will not\ndraw what the build provisions.\nEach entry is a `kind` plus a `name`, and you pick the kind by WHAT the target\nis. The kind-only fields below are exhaustive: one of them on another kind is a\nschema violation that both the zod write-gate and the Go fold gate reject.\n\n### The four kinds\n\n| `kind` | The target is | `name` comes from | Kind-only fields | Discover with |\n|---|---|---|---|---|\n| `component` | a SIBLING in this design that this component CALLS | the sibling's own name | — | this design |\n| `org-service` | a service ANOTHER project publishes for cross-project use | the provider's exact name, **copied verbatim** | — | `list_org_endpoints`, then `list_org_component_endpoints` |\n| `platform-resource` | a backing resource the platform provisions (database, cache, IDP) | **your choice** — it becomes the env-var prefix | `resourceType` (a registered type), `parameters` | `list_platform_resource_types` |\n| `external` | a system OUTSIDE the platform (a SaaS API, a legacy service) | a **Registered External resource**'s exact name, else a **new** name for a **Project External resource** | none on the component — the definition is the dependency's own file (below) | `list_external_resources` + `get_external_resource_schema`, else `web_search` |\n\n**Reuse.** Call that last column's tool before authoring the entry, and take\nthe name and schema from the matching row rather than from the requirement's\nwording — a **Registered External resource** described as \"transactional email\ndelivery\" is the right reuse for an \"email\" need even when its name (`sendgrid`)\ndoesn't echo the requirement. When several rows could fill the role, the\norg-level one wins (a Registered External over a new Project External name; an\norg-service over a sibling you would otherwise add; a listed cluster resource\ntype for `resourceType`). This step is done when every `external`,\n`org-service`, and `platform-resource` emitted this turn is taken from this\nturn's matching `list_*` result (exact `name` or `resourceType`), unless this\nturn is a user-asked reconsider. When nothing the catalog returns fills the\nrole, omit the entry and list the gap under **Needs your input** rather than\ncoining a name: a `resourceType` that was not in this turn's\n`list_platform_resource_types` result fails Build.\n\n```json\n\"dependencies\": [\n  { \"kind\": \"component\", \"name\": \"expense-api\" },\n  { \"kind\": \"platform-resource\", \"name\": \"orders-db\", \"resourceType\": \"postgres-cnpg\" },\n  { \"kind\": \"external\", \"name\": \"github\",\n    \"description\": \"Call GitHub issues + PRs for the sync story.\" }\n]\n```\n\nThe `github` entry is a REFERENCE: its definition — provider, style, contract,\nconfig keys — is `specs\u002Fdesign\u002Fdependencies\u002Fgithub\u002Fdependency.json`, written\nonce and shared by every component that uses it. A component's `description`\non the edge says why THIS component uses it; the dependency's own description\nsays what the system is.\n\n#### Reading a provider's real contract\n\nOnce you have an `org-service` provider's name, call\n`list_org_component_endpoints` and base that dependency's `description` on the\noperations its contract actually exposes:\n\n| `spec.availability` | Where the contract is |\n|---|---|\n| `inline` | `spec.inlineContent` IS the OpenAPI document — read it directly |\n| `repo` | the row's `owner`\u002F`repo`\u002F`subdir`\u002F`branch` locate the source: `search_remote_git_code` under that `subdir`, then `get_remote_git_file_contents` |\n| `none` | nothing resolvable — say so plainly in the `description`, never invent a shape |\n\n### Traps\n\n- **Never the reverse edge.** Declare a dependency ONLY on the caller, naming the\n  callee: a web-app depends on the API it calls; the API does not depend on the\n  web-app. A component this one doesn't call is not a dependency of it — don't\n  list one \"for reference\".\n- **A role is not a name.** The requirement says \"the organization's directory\n  service\"; the provider is usually called something else (`employee-service`).\n  The same for a `platform-resource`: the PRD names a capability, `resourceType`\n  is the catalog row's `name` from this turn's `list_platform_resource_types`.\n  Copy that `name`. Look it up — a name coined from the role or capability words\n  matches no provider or type and hard-fails the build.\n- **A `platform-resource`'s `name` becomes the env-var prefix** for every one of\n  its outputs (`orders-db` → `ORDERS_DB_HOST`, and for a SPA\n  `window._env_.ORDERS_DB_*`), so pick a clear one: renaming it later renames the\n  component's whole config surface.\n- **Secret-bearing dependencies belong on a `service`, not a\n  `web-application`.** A web-application ships to the browser, so anything it\n  holds is visible in dev tools — and *every* output of a `platform-resource` it\n  declares is emitted into `window._env_`, so a database dependency on a SPA\n  publishes its password. The secure default: attach the dependency to a backend\n  `service` and give the web-app a `component` edge to it, so the service proxies\n  and the SPA never sees the credential. A web-application may declare an\n  `external` dependency directly only when NONE of its `config` keys need\n  `secret: true` — a genuinely public API, or one the END USER authenticates with\n  their own in-browser OAuth. The one exception is `thunder-app` (next bullet):\n  its outputs are public OIDC client config by design. The schema does not reject\n  a secret on a web-application, so apply this as the architect's judgment call,\n  not a rule to route around.\n- **`thunder-app` is how sign-in happens, and nothing else provisions it.** When\n  the spec implies users sign in, declare it on BOTH the SPA and each protected\n  service under the SAME dependency `name` — that shared name is what ties\n  sign-in to token-carrying API calls. With no such dependency the SPA deploys\n  unable to sign in. `thunder-app` takes no `parameters` and **nothing about the\n  sign-in client is authored by hand**: its display name, its redirect URIs and\n  its scope list are all derived by the platform (the scopes from the permission\n  catalog `security-design` writes on `security.json`). A project with no web\n  application still declares the dependency on its API — the client is what the\n  platform's Try it app and the validation agent sign in through.\n  `thunder-authentication` owns the coding-time rule, and `security-design`\n  owns which roles sign in through it and what they may do.\n\n### Resolving an `external` dependency\n\n**One dependency, one definition.** An external dependency lives in its own\ndirectory, `specs\u002Fdesign\u002Fdependencies\u002F\u003Cname>\u002F`, and a component only\nreferences it by `name`. The write-gate refuses a component that names an\nexternal dependency whose directory has no `dependency.json` yet, so write\nthe dependency file BEFORE the component that uses it (a `declare_plan`\nentry per dependency keeps the rail honest). The name is the same\nidentifier everywhere: the directory, the cell's `south` node, the\ncomponent's reference, and — for a reuse — the org registry.\n\n`dependency.json` — ONE PROJECT'S USE OF A RESOURCE. The `resource` block is\nwhat the thing is, in the one shape a resource has everywhere (the org\nregistry holds the same shape); around it sit this project's provenance and\nopen suggestions:\n\n```json\n{\n  \"name\": \"payment-service\",\n  \"resource\": {\n    \"name\": \"payment-service\",\n    \"description\": \"Charges the customer for shipping once a box is priced.\",\n    \"provider\": \"Stripe\",\n    \"config\": [ { \"key\": \"PAYMENT_API_KEY\", \"secret\": true, \"description\": \"Stripe secret API key\" } ],\n    \"contract\": { \"type\": \"openapi\", \"path\": \"openapi.yaml\", \"origin\": \"provider\" }\n  },\n  \"provenance\": { \"sourceUrl\": \"https:\u002F\u002F…\u002Fspec3.json\" }\n}\n```\n\nThere is no `style`: how the component consumes the system is computed from\n`contract.type` (`openapi` → a REST client, `graphql` → a GraphQL client, `sdk`\n→ a vendor library). The contract is ONE WHOLE DOCUMENT beside the file, never\na slice — the coding agent cuts what its component calls at coding time — and\nit is always `{ \"type\", \"path\" }`, a file name in the directory: there is no\nURL form, and the coding agent must find nothing in the repo it could follow\noff it. `contract.origin` says where the file came from: `registry` (copied\nfrom the organization's record), `provider` (the provider's published\ndocument, fetched by the platform), `derived` (you wrote it from the\nprovider's developer reference), `assumed` (you wrote it from less, under the\nuser's authorization — `contract.accepted` is that record, and the platform\nwrites it, never you).\n\nWork each one in order. A design run does them in two passes:\n\n- **With the component that uses it** (the design's Architecture step):\n  steps 1, 2 and 5. The `dependency.json` of this pass says only what the\n  dependency is: `name`, and `resource.name`, `resource.description`, and\n  `resource.provider` or `suggestions`. No `web_search`, no `contract`, no\n  `config`, no `provenance`, no contract file, no `assumptions.md`.\n- **In the design's Dependencies step**, when the components show what they\n  call: steps 3 and 4 for a dependency with a provider. Edit its\n  `dependency.json` to add what they find, write the contract file when you\n  derive it, then write `assumptions.md` (**What the design assumes about a\n  dependency**, below).\n\nThe steps:\n\n1. **Reuse first.** Call `list_external_resources`. It lists the resources\n   the ORGANIZATION registered — never one another project defined for\n   itself. When a row fits, write the stub and nothing more:\n   `{ \"name\": \"\u003Cits exact name>\", \"resource\": { \"ref\": \"\u003Cits exact name>\", \"name\": \"\u003Cits exact name>\" } }`.\n   The platform fills the block at save — provider, config keys, the\n   organization's consumption instructions — and copies the record's contract\n   document beside the file (`origin: \"registry\"`). Never retype the keys or\n   the instructions: their identities are load-bearing, and a retyped key is\n   how they drift. The build collects no values for it — org values stay on\n   the registered name. Then stop. A user-asked reconsider may switch to a\n   different registered name, or define a **Project External resource**\n   inline under a new name.\n2. **The PRD names the service, or nobody has.** The user chooses providers;\n   you never do. Two outcomes, never a `status`:\n   - **The requirements' decisions name a service for this capability**\n     (\"Payments: Stripe\") → write `resource.provider`. Its contract is\n     step 3. An org or platform skill that mandates a vendor\n     counts the same way. Nothing else does: \"the requirement implies it\",\n     \"this one is popular\", \"there is only one real option\" are guesses, and\n     a guess is the user's to make.\n   - **No provider named** → write the NEED only: `name`, `resource.name`,\n     `resource.description`, and `suggestions` — providers commonly used for\n     this capability, from what you know, any number, each\n     `{ \"name\", \"style\"?, \"description\"? }` with the one distinction that\n     matters for THIS product. No `web_search`, no `provider`, no `contract`,\n     no `config` — the config keys follow the provider, and none is chosen.\n     The definition then offers **Select a provider**, which runs the\n     `resolve-dependency` flow: it asks the user, with your suggestions as\n     the options, and does the research. This is the EXPECTED outcome for a\n     choosable dependency; do not force a pick the PRD does not make.\n   The dependency IS the service the product needs, so name it\n   `\u003Ccapability>-service` (`currency-service`, `payment-service`,\n   `email-service`); the chosen system is its provider.\n3. **Get the contract document on disk — for a named provider only.** A\n   ladder, climbed in order:\n   - **Point at the published document.** `web_search` for the provider's\n     published OpenAPI or GraphQL document, then write, INSIDE `resource`,\n     `\"contract\": { \"type\": \"openapi\", \"path\": \"openapi.yaml\", \"origin\": \"provider\" }`\n     (or `graphql` \u002F `schema.graphql`), and `\"provenance\": { \"sourceUrl\": \"\u003Cits URL>\" }`\n     beside `resource` at the top level — and STOP. A `contract` written\n     anywhere but inside `resource` is refused. Do NOT fetch the document and\n     do NOT `addFile` it: the\n     platform fetches it at save (https only, at most 5 MiB), lands it beside\n     the definition and fills the hash. A document a user hands over goes\n     through **Provide interface** on the definition and lands the same way.\n   - **Derive** it from the provider's OWN developer reference when it names\n     every operation the design calls with parameters and responses:\n     `addFile` an `openapi.yaml` with an `x-aep-source: \u003Cpage>` on every\n     operation, `resource.contract.origin: \"derived\"`, `provenance.sourceUrl` =\n     the reference's root page. That needs no permission: the dependency reads\n     resolved, flagged *derived*.\n   - With no such documentation either, never guess during the design turn:\n     leave `contract` unset, say so under **Needs your input**, and the\n     `resolve-dependency` flow takes it from there (it may write an ASSUMED\n     contract, but only under the user's authorization).\n   An SDK is a contract of its own type: `{ \"type\": \"sdk\", \"path\": \"sdk.json\" }`,\n   and `sdk.json` names the package per implementation language. This rung\n   is for a NAMED provider only — an open capability gets no interface\n   search at all.\n4. **Derive `resource.config` keys** from the contract — a REST API's\n   `components.securitySchemes`, an SDK's auth documentation — for a named\n   provider only; a definition with no provider carries no keys (the gate\n   refuses them). A copied registered resource already carries its keys —\n   keep them.\n5. **Reference it** from each consuming component:\n   `{ \"kind\": \"external\", \"name\": \"\u003Cname>\", \"description\": \"\u003Cwhy this component uses it>\" }`.\n   `style`, `package`, `specPath`, `suggestions`, `config` on the component\n   are refused — they belong in the dependency file.\n\n### What the design assumes about a dependency\n\nThe design can rely on a fact about an outside service that nothing confirms:\nnot the contract, not the provider's documentation, not the requirements.\nWrite each such fact in `specs\u002Fdesign\u002Fdependencies\u002F\u003Cname>\u002Fassumptions.md`,\nbeside the dependency's definition and contract, in the design's Dependencies\nstep:\n\n```markdown\n# What the design assumes about Xero\n\n- Xero accepts up to 500 bill lines in one batch. *assumed* (F3)\n- Xero answers a duplicate bill number with an error, not a second bill. *assumed* (F3)\n```\n\n- One line per fact, in one sentence. Then `*assumed*`. Then, in parentheses,\n  the IDs of the features that rely on it.\n- A fact the contract states, or a requirement line with its source (`[Xero\n  API docs]`), is not an assumption: do not repeat it here.\n- A choice about the project's own system (\"the export retries 3 times\") is\n  a technical decision, not an assumption. It goes in\n  `specs\u002Fdesign\u002Fdecisions.md` (`design`).\n- Write no file when the design assumes nothing about the service.\n- Keep each line that still holds on a later run. The tag comes off when\n  someone confirms the fact.\n\n### Config-key conventions\n\n`resource.config` is the env-var schema the consuming component codes against. Use\n`SCREAMING_SNAKE_CASE` keys and keep them minimal — only what the component\nreads. `secret` is opt-in: set `\"secret\": true` ONLY for credentials (they route\nthrough the secret path), and OMIT it entirely otherwise. Give each key a\n`description` saying what the value is and where the user finds it\n(`{ \"key\": \"STRIPE_API_KEY\", \"secret\": true, \"description\": \"Your Stripe secret\nAPI key\" }`) — the Build dependency drawer shows it under the field. For a\nNON-secret key whose sensible default you can infer, add `defaultValue` and the\ndrawer pre-fills it (`{ \"key\": \"AWS_REGION\", \"defaultValue\": \"us-east-1\" }`).\nNEVER set `defaultValue` on a secret — a credential has no default to invent. An\n`external` dependency should almost always carry at least one key: the\nvalue-collection gate needs something to collect.\n\n#### How the platform derives status\u002Freason\n\nYou never author `status`\u002F`reason`. The platform reads them off the dependency\nfile at read time, plus one registry lookup for a copy, first match wins:\n\n1. `resource.ref` set and the organization has a registered resource of that\n   name → the copy stands; only the contract is checked (rule 4)\n2. `resource.ref` set and NO registered resource of that name → `unresolved`\u002F\n   `needs-input` — the definition says \"The organization has no registered\n   resource with this name\" and offers Select a provider\n3. no `resource.provider` (`suggestions` open or not) → `unresolved`\u002F\n   `needs-input` — the user has not chosen a service; the definition asks them\n4. no `resource.contract`, or a contract whose file is not on disk (the\n   platform's fetch may still be owed or may have failed) → `unresolved`\u002F\n   `needs-contract`\n5. `contract.origin: \"assumed\"` with no `accepted` → `unresolved`\u002F\n   `needs-acceptance`\n6. otherwise → `resolved` — flagged `registered` for a copy, `assumed` when\n   the contract was accepted as an assumption, `derived` when it was written\n   from the provider's own documentation, `sdk-only` for an sdk contract,\n   `stale` when a copy's document no longer matches the organization's\n\n`component` is always `resolved` here. A `platform-resource` is too — once\nemitted — so only emit one whose `resourceType` is a `name` from this turn's\n`list_platform_resource_types`. An `org-service` resolves on catalog visibility,\nand is `blocked`\u002F`access-required` when the provider exists but this project\ncannot see it. `needsSpec`, `specPath` and the other definition fields are gone from the\ncomponent's schema — a draft carrying them fails the write-gate with a message\nnaming the dependency file they moved to.\n\n### Narrating the design turn\n\nThe design-generate turn runs in the chat panel, so your turn text is what the\nuser watches live. **Narrate each dependency decision in one plain-prose line as\nyou settle it** in the Dependencies step, before moving to the next:\n\n- resolved → `✓ \u003Ccapability>: using \u003Cchoice>` (say `, document fetched at\n  save` when you pointed at a published one, `, interface derived from docs`\n  when you wrote it from the provider's reference, `, registered — the\n  organization's resource, copied here` for a reuse)\n- needs-input → `\u003Ccapability>: your choice — A \u002F B \u002F C are common; select a\n  provider on its definition`\n- needs-contract → `\u003Ccapability>: \u003Cprovider> chosen, no published contract\n  found — you can upload one or let me assume it, from the dependency's definition in the spec view`\n\nNever block the design on an unresolved dependency — print the line\nand keep emitting the rest; the user replies in the same chat to steer it, now or\nlater. Then close as `design` says. Its **\"Needs your input\"** block lists ONLY\nthe dependencies still unresolved, each as a LINK to its definition —\n`[currency-service](aep:\u002F\u002Fspec\u002Fspecs\u002Fdesign\u002Fdependencies\u002Fcurrency-service\u002Fdependency.json)`\n— followed by the single thing you need (the console opens the definition\nfrom the link; this link form is the one place a repo path is allowed in\nyour prose).\nEach **Needs your input** line names the dependency the way its definition in the spec view does, so\nthe user can click through and press **Resolve** — that runs the\n`resolve-dependency` flow, which is where the contract gets provided or\nassumed; the design turn never waits for it.\n\n### Resolving or reconsidering a named dependency on request\n\n`\u002Fresolve-dependency \u003Cname>` runs the `resolve-dependency` skill — the guided\nflow that takes one dependency from open to resolved; read that skill when the\ninstruction names it. A plain chat turn may still point you at a dependency\n(\"reconsider `stripe`\"): read `specs\u002Fdesign\u002Fdependencies\u002F\u003Cname>\u002Fdependency.json`\nfrom the snapshot and act on its current state.\n\n- **No service chosen — it carries `suggestions`, or nothing.** The user's\n  answer, if the instruction carries one (a service name or a document URL),\n  IS the choice: set `provider` + `style`, remove `suggestions`, and go for\n  the contract (step 3). With no answer, put the choice to them — the\n  suggestions, or what your research finds, each with a one-line distinction,\n  plus that they may name another — and write nothing until they answer.\n- **Already resolved — reconsider.** This is the only branch that may leave a\n  catalog row that still fills the role. Present fresh alternatives in the\n  conversation, or repin to the Registered name the user picks, or start a\n  **Project External resource** under a **new** name.\n\nEdit ONLY that dependency's file: re-emit the whole `dependency.json` (never a\npatch) with every field you are not changing carried over exactly. The\ncomponents' references do not change. Never write or alter `assumed` — that is\nthe user's record; the write-gate refuses it.\n\n### Descriptions, and the per-component artifacts\n\nEvery dependency carries a one-line `description`: what the target is and how\nthis component uses it. Source it per kind — an `external`'s definition file\ncarries what the system is (the consumption instructions when the name is a\nRegistered External resource) and the component's reference why this component\ncalls it; an `org-service`'s\nsays the specific\noperations it calls from the discovered contract, or plainly that no contract was\nresolvable, never a guess; a `platform-resource`'s says what it stores. The\nconsole shows it in the dependency drawer and the coding agent relies on it to\nintegrate correctly.\n\nOne component per directory. Every `web-application` gets a prototype,\n`prototype.json` and `prototype.tsx` (`prototype` governs it); every `service`\ngets an `openapi.yaml` (`openapi-conventions` governs it); every `ai-agent`\ngets an `agent.afm.md` (`agent-building` governs it). `design` says when each\nis written. Other kinds (scheduled tasks, workers, …) carry no extra artifact\nyet — capture their behaviour fully in `description` and `dependencies`.\n","# Architecture\n\nComponent decomposition and per-component facts. The cell (`cell-design`\nskill) declares WHAT exists; this skill owns deciding the decomposition and\nENRICHING each component's `design.json`.\n\n## Scaffold first, enrich second\n\nWhen the cell is saved, the platform scaffolds\n`specs\u002Fdesign\u002Fcomponents\u002F\u003Cid>\u002Fdesign.json` for every deployable component —\nthe mechanical fields (name, type, version, buildpack, appPath, entrypoint, a\ndefault exposure) plus a `\"language\": \"TBD\"` sentinel. Your job is the\nJUDGMENT fields, and the build gate refuses a deployable component left\nunenriched:\n\n- **language** — **Language is decided in this order**: the organization skill's Tech stack\ndefault first, then a language the requirements name, then the platform\ndefault — **Ballerina** for backend services, TypeScript for web apps.\nReplace the scaffold's `\"language\": \"TBD\"` sentinel with the decided value;\nthe build gate refuses a component whose language is still TBD. That choice is a fact\nyou record, not a preference you re-derive per component: write it as the\ncomponent's `language` and pin the matching stack skill in `skillsPinned`.\n- **stories** — the IDs of the stories this component serves, as written in\n  the feature files (e.g. `\"stories\": [\"F1.1\", \"F1.2\", \"F2.1\"]`). Claim every\n  story the component actually serves: the build gate refuses the tag while\n  any story is claimed by no component, or an ID the requirements do not\n  define is cited.\n- **dependencies** — the playbook below.\n- **description** — one paragraph: single responsibility, port\u002Fentrypoint\n  expectations, and what it explicitly does NOT do.\n- **skillsPinned** — After emitting or changing a component's design, record the skills that\ncomponent's build actually needs as a `skillsPinned` array **inside that\ncomponent's `specs\u002Fdesign\u002Fcomponents\u002F\u003Cname>\u002Fdesign.json`** — use the exact\ncatalog names, e.g. a Ballerina API service →\n`[\"openapi-conventions\", \"ballerina\"]` (a Go one → `[\"openapi-conventions\",\n\"go\"]`); a web-application → `[\"react-webapp\", \u003Cdesign system>]`,\nwhere `\u003Cdesign system>` is the skill name in the **UI design system** section of\nthe Organization defaults block in your instructions. A web app's UI toolkit is\na settled organization decision, so pin it on **every** web-application rather\nthan leaving it to a description to trigger — and read the name from that\nsection every time instead of remembering one, because it differs per\norganization. If that section is absent or empty, pin only `react-webapp`.\nNever substitute a design system the organization defaults do not name.\n\n**Pin the design system; do not consult it.** A design system is built against,\nnot designed with — it is the coding run's to load, and its theming is a settled\norganization decision that no design-time question reopens. Never ask the user\nabout colors, themes, or look and feel. An ai-agent → `[\"agent-building\"]`. **Pin the auth skills the same way,\nnever leave them to a description:** `\"thunder-authentication\"` on **both**\nsides of sign-in — every component that declares the `thunder-app` dependency,\nthe SPA *and* each protected backend it calls — and `\"api-management\"` on every\nservice that sits behind the gateway. The first owns the SPA's sign-in, its\nscreen gates and the backend's verification of the gateway's assertion; the\nsecond owns the gateway's contract. It is a JSON key on the component's design\nobject, so include it when you write that `design.json` (addFile\u002FeditFile) —\n`design.json` is its only home.\nEach component carries only the skills its own build needs.\n\nWriting the whole enriched file yourself (removeFile + addFile with every\nfield) is equally valid — the scaffold is a safety net, not a required\nintermediate.\n\n## Deriving components — deployment units the requirements justify\n\nA component is one independently deployable unit, NOT a domain concept. The\nright number comes from the requirements: for every component you must be\nable to say \"this deploys and evolves independently because \u003Csomething the\nrequirements state>\". Write that justification into the component's\n`description`.\n\nA requirement justifies a SEPARATE component when it shows:\n\n- a distinct user-facing surface — e.g. an internal admin portal AND a\n  customer-facing app with different users and lifecycles → two web-applications;\n- a genuinely different runtime or scaling profile — e.g. an async\n  worker\u002Fbatch processor beside an interactive API, or a long-running\n  AI\u002Finference service;\n- a technology the rest of the system doesn't share — e.g. a Python ML\n  service beside a Ballerina API;\n- an explicitly separate lifecycle or ownership stated in the requirements.\n\nDo NOT split by:\n\n- entity or domain concept — claims-service, users-service,\n  receipts-service... is a domain model dressed as a topology; those are\n  modules of ONE service;\n- layer — auth, notifications, file storage as own services when they are\n  modules of the API; a single frontend split into `ui-shell` + `ui-pages` when\n  it is one bundle;\n- infrastructure — api-gateway, database, queue, auth server are NEVER\n  components; the platform provides them;\n- periodic work — a nightly digest or a cleanup sweep is a background task the\n  owning service starts, not a component of its own. Only a genuinely different\n  runtime or scaling profile (above) justifies splitting one off.\n\n**Never invent a user-facing surface the requirements did not ask for.** A\nsystem whose requirements describe a browser app lands at one service + one\nweb-application; a system whose only surface is an agent lands at one service +\none ai-agent and NO web-application. An agent is a complete surface on its own —\nthe platform's Try it app talks to a deployed agent directly — so \"there must be\nsomething for the user to open\" is not a reason to add a SPA. Adding an\nunrequested web-application is the same failure as splitting by domain concept:\na shape imposed on the requirements instead of read from them. Name components\nin kebab-case after their responsibility (`expense-api`, `expense-webapp`,\n`report-worker`, `packing-agent`).\n\n**An AI agent is `\"ai-agent\"`, and every model call is one.** It is the only\ncomponent type with model access. Reach for it when the requirements call for a\nconversational or autonomous surface — a user talking to the system in their own\nwords rather than filling in a form — and for every agent the requirements'\ndecisions name. Its behaviour is authored as\n`agent.afm.md` (the `agent-building` skill), it is implemented in TypeScript, and\nit pins `[\"agent-building\"]`. It is a normal deployable that calls other\ncomponents over HTTP: give it a `component` dependency for every API it uses.\n**An agent a signed-in user reaches is a protected backend** — whether the\ncaller is a sibling web-application or the platform's Try it app, set\n`\"exposure\": \"internet\"` (a browser cannot reach an intranet address, so an\nintranet agent has nothing that can talk to it) AND give it the project's\nshared `thunder-app`\ndependency, under the same dependency NAME the SPA and the sibling APIs use,\nexactly as you would for a service. It is the same sign-in and the same OAuth\napp, not a second one. Do this even when the agent stores nothing and every API\nbehind it already authorises: a service you can call without a token costs CPU,\nan agent you can call without a token costs money on the organisation's model\nkey, so an unauthenticated agent endpoint is a billing hole rather than a\ntolerable one.\n**Declare no dependency for model access** — every `ai-agent` gets it from its\ncomponent type, on the organisation's own key, so there is nothing to choose.\n**An agent whose work needs files declares them.** Give it `x-aep.attachments`\nwith the types and counts the feature needs (`agent-building` has the ceilings\nand the contract). One whose work runs with nobody signed in — a ticket from an\narriving email, a nightly job — does not fit yet: ask the user at design\naltitude whether a person does that step for now or the feature waits.\n**An `ai-agent` with server memory needs a Postgres for its conversation\nstore.** Give it a `platform-resource` dependency with `resourceType:\n\"postgres-cnpg\"` — the PVC-backed type, so a conversation survives a pod\nrestart; the older ephemeral `postgres` type is not installed and referencing\nit fails provisioning. Dedicated is the default — `{ \"kind\":\n\"platform-resource\", \"name\": \"memory-db\", \"resourceType\": \"postgres-cnpg\" }`,\ninjected as `MEMORY_DB_HOST` \u002F `MEMORY_DB_PORT` \u002F `MEMORY_DB_DBNAME` \u002F\n`MEMORY_DB_USER` \u002F `MEMORY_DB_PASSWORD` (`postgres-cnpg` has no `url` output;\nthe generic platform-resource wiring injects each of its five outputs as\n`\u003CDEP_NAME>_\u003COUTPUT>`, uppercased). When the project already carries a\nPostgres and the user prefers one instance, declare the SAME dependency name\nthe sibling service uses — same-name resolution to one shared instance is the\n`thunder-app` sharing rule, and it applies here too. Either way the agent owns\nits `conversations` table exclusively: no other component touches it, and the\nagent touches nothing else in a shared instance. Business data is still not\nthe agent's to hold — anything beyond its own conversation belongs behind a\nservice that owns it.\n\n**Component `type` is a fixed vocabulary — use the EXACT string.** A backend is\n`\"service\"`; a browser app is `\"web-application\"` (OpenChoreo's own term). Write\n`\"web-application\"` verbatim — NOT `\"webapp\"`, `\"web-app\"`, or `\"webApplication\"`\n(those are rejected, and a wrong value silently breaks the app's deployment and\nruntime config). The `-webapp` in a component NAME is fine; the `type` is still\n`\"web-application\"`. Other kinds the requirements imply (`\"scheduled-task\"`,\n`\"worker\"`, …) are captured verbatim, but the platform installs component types\nfor `service` and `web-application` ONLY — anything else records intent and does\nnot deploy, so reach for one when the requirements truly force it, never as the\ndefault home for periodic work.\n\n## Per-component design.json\n\nEach component's structured facts live in ONE JSON document (no markdown, no\nfrontmatter). The platform validates each write against this schema and rejects\nviolations:\n\n```json\n{\n  \"name\": \"expense-api\",              \u002F\u002F MUST equal the directory name\n  \"type\": \"service\",                  \u002F\u002F EXACT kind: \"service\" or \"web-application\" (NEVER \"webapp\"\u002F\"web-app\"), or another the requirements imply (\"scheduled-task\", \"worker\", ...)\n  \"version\": \"0.1.0\",                 \u002F\u002F semantic version; 0.1.0 for a new component\n  \"language\": \"Ballerina\",            \u002F\u002F implementation language — \"Ballerina\" for a service unless the requirements say otherwise; \"TypeScript\" for a web-application\n  \"buildpack\": \"docker\",              \u002F\u002F always \"docker\"\n  \"appPath\": \"expense-api\",           \u002F\u002F repo-relative source dir — the component name\n  \"entrypoint\": \"deployment\u002Fservice\", \u002F\u002F deploy entry — PAIRS with `type`: \"deployment\u002Fservice\" for a service, \"deployment\u002Fweb-application\" for a web-application, \"deployment\u002Fai-agent\" for an ai-agent\n  \"exposure\": \"internet\",             \u002F\u002F \"internet\" (public) | \"intranet\" (internal only)\n  \"dependencies\": [ \u002F* see below — every dependency edge touching this component appears here *\u002F ],\n  \"description\": \"One paragraph: single responsibility, port\u002Fentrypoint expectations, and what it explicitly does NOT do.\",\n  \"stories\": [\"F1.1\", \"F1.2\", \"F2.1\"], \u002F\u002F story IDs THIS component serves — the build gate refuses the tag while any story is claimed by nobody\n  \"skillsPinned\": [\"openapi-conventions\", \"ballerina\"], \u002F\u002F the skills this component's build needs — see the field above\n  \"endpoint\": { \"name\": \"http\" } \u002F\u002F optional; see below\n}\n```\n\n`name`, `type`, `version`, `language`, `buildpack`, `appPath`, `entrypoint`,\n`exposure`, `description`, `dependencies`, `stories` and `skillsPinned` are\nrequired — `stories` and `skillsPinned` are as required as the rest, and a\ncomponent missing either fails the build gate, so emit them in the SAME write\nrather than as a follow-up edit. To CHANGE a\ndesign.json, re-emit the whole corrected file (removeFile + addFile) — never\npatch JSON with anchored edits. On INVALID_JSON or SCHEMA_VIOLATION, fix what\nthe message lists and re-emit.\n\n`endpoint` is optional: omit it and a service's endpoint takes the default name\n`\"http\"`. Declare `{ \"name\": \"\u003Cendpoint-name>\" }` only when the endpoint must be\nnamed otherwise — `name` is the single source of truth the coding agent copies\ninto `workload.yaml` and the managed-API gateway binds to. The port lives in\n`workload.yaml`, not here.\n\n**Platform-owned fields you never author**, in two kinds:\n\n- **Preserved verbatim** where the platform has already written them:\n  `exposesAPI`, `componentAgentInstructions`, and any dependency\n  `status`\u002F`reason`.\n- **Recomputed and overwritten** on every save: a dependency's `wiring` object.\n  The platform derives its `ref` and its env-var names from the dependency's name\n  and its resource type's declared outputs, so anything you write there is\n  discarded. `stories` is NOT in this class — it is yours to author, per the\n  **stories** field above.\n\n### dependencies — one entry per design.cell edge\n\n`dependencies` mirrors the DEPENDENCY edges of design.cell: every arrow\nbetween this component and another node appears here and vice versa — a\nmismatch is a defect. The `north ->` \u002F `west ->` gateway arrows are\n**exposure**, not dependencies: they live in this component's `exposure`\nfield and never become entries here. The platform enforces the membership\nhalf as you write: a dependency whose `name` is not a node the cell declares\nis refused (`UNKNOWN_DEPENDENCY`) with the cell statement to add. A resource\nyou discover here — a database, a cache — is a cell node too: `editFile`\n`specs\u002Fdesign\u002Fdesign.cell` to add `component \u003Cname> as \"…\" database` inside\nthe cell before the design.json that depends on it, or the diagram will not\ndraw what the build provisions.\nEach entry is a `kind` plus a `name`, and you pick the kind by WHAT the target\nis. The kind-only fields below are exhaustive: one of them on another kind is a\nschema violation that both the zod write-gate and the Go fold gate reject.\n\n### The four kinds\n\n| `kind` | The target is | `name` comes from | Kind-only fields | Discover with |\n|---|---|---|---|---|\n| `component` | a SIBLING in this design that this component CALLS | the sibling's own name | — | this design |\n| `org-service` | a service ANOTHER project publishes for cross-project use | the provider's exact name, **copied verbatim** | — | `list_org_endpoints`, then `list_org_component_endpoints` |\n| `platform-resource` | a backing resource the platform provisions (database, cache, IDP) | **your choice** — it becomes the env-var prefix | `resourceType` (a registered type), `parameters` | `list_platform_resource_types` |\n| `external` | a system OUTSIDE the platform (a SaaS API, a legacy service) | a **Registered External resource**'s exact name, else a **new** name for a **Project External resource** | none on the component — the definition is the dependency's own file (below) | `list_external_resources` + `get_external_resource_schema`, else `web_search` |\n\n**Reuse.** Call that last column's tool before authoring the entry, and take\nthe name and schema from the matching row rather than from the requirement's\nwording — a **Registered External resource** described as \"transactional email\ndelivery\" is the right reuse for an \"email\" need even when its name (`sendgrid`)\ndoesn't echo the requirement. When several rows could fill the role, the\norg-level one wins (a Registered External over a new Project External name; an\norg-service over a sibling you would otherwise add; a listed cluster resource\ntype for `resourceType`). This step is done when every `external`,\n`org-service`, and `platform-resource` emitted this turn is taken from this\nturn's matching `list_*` result (exact `name` or `resourceType`), unless this\nturn is a user-asked reconsider. When nothing the catalog returns fills the\nrole, omit the entry and list the gap under **Needs your input** rather than\ncoining a name: a `resourceType` that was not in this turn's\n`list_platform_resource_types` result fails Build.\n\n```json\n\"dependencies\": [\n  { \"kind\": \"component\", \"name\": \"expense-api\" },\n  { \"kind\": \"platform-resource\", \"name\": \"orders-db\", \"resourceType\": \"postgres-cnpg\" },\n  { \"kind\": \"external\", \"name\": \"github\",\n    \"description\": \"Call GitHub issues + PRs for the sync story.\" }\n]\n```\n\nThe `github` entry is a REFERENCE: its definition — provider, style, contract,\nconfig keys — is `specs\u002Fdesign\u002Fdependencies\u002Fgithub\u002Fdependency.json`, written\nonce and shared by every component that uses it. A component's `description`\non the edge says why THIS component uses it; the dependency's own description\nsays what the system is.\n\n#### Reading a provider's real contract\n\nOnce you have an `org-service` provider's name, call\n`list_org_component_endpoints` and base that dependency's `description` on the\noperations its contract actually exposes:\n\n| `spec.availability` | Where the contract is |\n|---|---|\n| `inline` | `spec.inlineContent` IS the OpenAPI document — read it directly |\n| `repo` | the row's `owner`\u002F`repo`\u002F`subdir`\u002F`branch` locate the source: `search_remote_git_code` under that `subdir`, then `get_remote_git_file_contents` |\n| `none` | nothing resolvable — say so plainly in the `description`, never invent a shape |\n\n### Traps\n\n- **Never the reverse edge.** Declare a dependency ONLY on the caller, naming the\n  callee: a web-app depends on the API it calls; the API does not depend on the\n  web-app. A component this one doesn't call is not a dependency of it — don't\n  list one \"for reference\".\n- **A role is not a name.** The requirement says \"the organization's directory\n  service\"; the provider is usually called something else (`employee-service`).\n  The same for a `platform-resource`: the PRD names a capability, `resourceType`\n  is the catalog row's `name` from this turn's `list_platform_resource_types`.\n  Copy that `name`. Look it up — a name coined from the role or capability words\n  matches no provider or type and hard-fails the build.\n- **A `platform-resource`'s `name` becomes the env-var prefix** for every one of\n  its outputs (`orders-db` → `ORDERS_DB_HOST`, and for a SPA\n  `window._env_.ORDERS_DB_*`), so pick a clear one: renaming it later renames the\n  component's whole config surface.\n- **Secret-bearing dependencies belong on a `service`, not a\n  `web-application`.** A web-application ships to the browser, so anything it\n  holds is visible in dev tools — and *every* output of a `platform-resource` it\n  declares is emitted into `window._env_`, so a database dependency on a SPA\n  publishes its password. The secure default: attach the dependency to a backend\n  `service` and give the web-app a `component` edge to it, so the service proxies\n  and the SPA never sees the credential. A web-application may declare an\n  `external` dependency directly only when NONE of its `config` keys need\n  `secret: true` — a genuinely public API, or one the END USER authenticates with\n  their own in-browser OAuth. The one exception is `thunder-app` (next bullet):\n  its outputs are public OIDC client config by design. The schema does not reject\n  a secret on a web-application, so apply this as the architect's judgment call,\n  not a rule to route around.\n- **`thunder-app` is how sign-in happens, and nothing else provisions it.** When\n  the spec implies users sign in, declare it on BOTH the SPA and each protected\n  service under the SAME dependency `name` — that shared name is what ties\n  sign-in to token-carrying API calls. With no such dependency the SPA deploys\n  unable to sign in. `thunder-app` takes no `parameters` and **nothing about the\n  sign-in client is authored by hand**: its display name, its redirect URIs and\n  its scope list are all derived by the platform (the scopes from the permission\n  catalog `security-design` writes on `security.json`). A project with no web\n  application still declares the dependency on its API — the client is what the\n  platform's Try it app and the validation agent sign in through.\n  `thunder-authentication` owns the coding-time rule, and `security-design`\n  owns which roles sign in through it and what they may do.\n\n### Resolving an `external` dependency\n\n**One dependency, one definition.** An external dependency lives in its own\ndirectory, `specs\u002Fdesign\u002Fdependencies\u002F\u003Cname>\u002F`, and a component only\nreferences it by `name`. The write-gate refuses a component that names an\nexternal dependency whose directory has no `dependency.json` yet, so write\nthe dependency file BEFORE the component that uses it (a `declare_plan`\nentry per dependency keeps the rail honest). The name is the same\nidentifier everywhere: the directory, the cell's `south` node, the\ncomponent's reference, and — for a reuse — the org registry.\n\n`dependency.json` — ONE PROJECT'S USE OF A RESOURCE. The `resource` block is\nwhat the thing is, in the one shape a resource has everywhere (the org\nregistry holds the same shape); around it sit this project's provenance and\nopen suggestions:\n\n```json\n{\n  \"name\": \"payment-service\",\n  \"resource\": {\n    \"name\": \"payment-service\",\n    \"description\": \"Charges the customer for shipping once a box is priced.\",\n    \"provider\": \"Stripe\",\n    \"config\": [ { \"key\": \"PAYMENT_API_KEY\", \"secret\": true, \"description\": \"Stripe secret API key\" } ],\n    \"contract\": { \"type\": \"openapi\", \"path\": \"openapi.yaml\", \"origin\": \"provider\" }\n  },\n  \"provenance\": { \"sourceUrl\": \"https:\u002F\u002F…\u002Fspec3.json\" }\n}\n```\n\nThere is no `style`: how the component consumes the system is computed from\n`contract.type` (`openapi` → a REST client, `graphql` → a GraphQL client, `sdk`\n→ a vendor library). The contract is ONE WHOLE DOCUMENT beside the file, never\na slice — the coding agent cuts what its component calls at coding time — and\nit is always `{ \"type\", \"path\" }`, a file name in the directory: there is no\nURL form, and the coding agent must find nothing in the repo it could follow\noff it. `contract.origin` says where the file came from: `registry` (copied\nfrom the organization's record), `provider` (the provider's published\ndocument, fetched by the platform), `derived` (you wrote it from the\nprovider's developer reference), `assumed` (you wrote it from less, under the\nuser's authorization — `contract.accepted` is that record, and the platform\nwrites it, never you).\n\nWork each one in order. A design run does them in two passes:\n\n- **With the component that uses it** (the design's Architecture step):\n  steps 1, 2 and 5. The `dependency.json` of this pass says only what the\n  dependency is: `name`, and `resource.name`, `resource.description`, and\n  `resource.provider` or `suggestions`. No `web_search`, no `contract`, no\n  `config`, no `provenance`, no contract file, no `assumptions.md`.\n- **In the design's Dependencies step**, when the components show what they\n  call: steps 3 and 4 for a dependency with a provider. Edit its\n  `dependency.json` to add what they find, write the contract file when you\n  derive it, then write `assumptions.md` (**What the design assumes about a\n  dependency**, below).\n\nThe steps:\n\n1. **Reuse first.** Call `list_external_resources`. It lists the resources\n   the ORGANIZATION registered — never one another project defined for\n   itself. When a row fits, write the stub and nothing more:\n   `{ \"name\": \"\u003Cits exact name>\", \"resource\": { \"ref\": \"\u003Cits exact name>\", \"name\": \"\u003Cits exact name>\" } }`.\n   The platform fills the block at save — provider, config keys, the\n   organization's consumption instructions — and copies the record's contract\n   document beside the file (`origin: \"registry\"`). Never retype the keys or\n   the instructions: their identities are load-bearing, and a retyped key is\n   how they drift. The build collects no values for it — org values stay on\n   the registered name. Then stop. A user-asked reconsider may switch to a\n   different registered name, or define a **Project External resource**\n   inline under a new name.\n2. **The PRD names the service, or nobody has.** The user chooses providers;\n   you never do. Two outcomes, never a `status`:\n   - **The requirements' decisions name a service for this capability**\n     (\"Payments: Stripe\") → write `resource.provider`. Its contract is\n     step 3. An org or platform skill that mandates a vendor\n     counts the same way. Nothing else does: \"the requirement implies it\",\n     \"this one is popular\", \"there is only one real option\" are guesses, and\n     a guess is the user's to make.\n   - **No provider named** → write the NEED only: `name`, `resource.name`,\n     `resource.description`, and `suggestions` — providers commonly used for\n     this capability, from what you know, any number, each\n     `{ \"name\", \"style\"?, \"description\"? }` with the one distinction that\n     matters for THIS product. No `web_search`, no `provider`, no `contract`,\n     no `config` — the config keys follow the provider, and none is chosen.\n     The definition then offers **Select a provider**, which runs the\n     `resolve-dependency` flow: it asks the user, with your suggestions as\n     the options, and does the research. This is the EXPECTED outcome for a\n     choosable dependency; do not force a pick the PRD does not make.\n   The dependency IS the service the product needs, so name it\n   `\u003Ccapability>-service` (`currency-service`, `payment-service`,\n   `email-service`); the chosen system is its provider.\n3. **Get the contract document on disk — for a named provider only.** A\n   ladder, climbed in order:\n   - **Point at the published document.** `web_search` for the provider's\n     published OpenAPI or GraphQL document, then write, INSIDE `resource`,\n     `\"contract\": { \"type\": \"openapi\", \"path\": \"openapi.yaml\", \"origin\": \"provider\" }`\n     (or `graphql` \u002F `schema.graphql`), and `\"provenance\": { \"sourceUrl\": \"\u003Cits URL>\" }`\n     beside `resource` at the top level — and STOP. A `contract` written\n     anywhere but inside `resource` is refused. Do NOT fetch the document and\n     do NOT `addFile` it: the\n     platform fetches it at save (https only, at most 5 MiB), lands it beside\n     the definition and fills the hash. A document a user hands over goes\n     through **Provide interface** on the definition and lands the same way.\n   - **Derive** it from the provider's OWN developer reference when it names\n     every operation the design calls with parameters and responses:\n     `addFile` an `openapi.yaml` with an `x-aep-source: \u003Cpage>` on every\n     operation, `resource.contract.origin: \"derived\"`, `provenance.sourceUrl` =\n     the reference's root page. That needs no permission: the dependency reads\n     resolved, flagged *derived*.\n   - With no such documentation either, never guess during the design turn:\n     leave `contract` unset, say so under **Needs your input**, and the\n     `resolve-dependency` flow takes it from there (it may write an ASSUMED\n     contract, but only under the user's authorization).\n   An SDK is a contract of its own type: `{ \"type\": \"sdk\", \"path\": \"sdk.json\" }`,\n   and `sdk.json` names the package per implementation language. This rung\n   is for a NAMED provider only — an open capability gets no interface\n   search at all.\n4. **Derive `resource.config` keys** from the contract — a REST API's\n   `components.securitySchemes`, an SDK's auth documentation — for a named\n   provider only; a definition with no provider carries no keys (the gate\n   refuses them). A copied registered resource already carries its keys —\n   keep them.\n5. **Reference it** from each consuming component:\n   `{ \"kind\": \"external\", \"name\": \"\u003Cname>\", \"description\": \"\u003Cwhy this component uses it>\" }`.\n   `style`, `package`, `specPath`, `suggestions`, `config` on the component\n   are refused — they belong in the dependency file.\n\n### What the design assumes about a dependency\n\nThe design can rely on a fact about an outside service that nothing confirms:\nnot the contract, not the provider's documentation, not the requirements.\nWrite each such fact in `specs\u002Fdesign\u002Fdependencies\u002F\u003Cname>\u002Fassumptions.md`,\nbeside the dependency's definition and contract, in the design's Dependencies\nstep:\n\n```markdown\n# What the design assumes about Xero\n\n- Xero accepts up to 500 bill lines in one batch. *assumed* (F3)\n- Xero answers a duplicate bill number with an error, not a second bill. *assumed* (F3)\n```\n\n- One line per fact, in one sentence. Then `*assumed*`. Then, in parentheses,\n  the IDs of the features that rely on it.\n- A fact the contract states, or a requirement line with its source (`[Xero\n  API docs]`), is not an assumption: do not repeat it here.\n- A choice about the project's own system (\"the export retries 3 times\") is\n  a technical decision, not an assumption. It goes in\n  `specs\u002Fdesign\u002Fdecisions.md` (`design`).\n- Write no file when the design assumes nothing about the service.\n- Keep each line that still holds on a later run. The tag comes off when\n  someone confirms the fact.\n\n### Config-key conventions\n\n`resource.config` is the env-var schema the consuming component codes against. Use\n`SCREAMING_SNAKE_CASE` keys and keep them minimal — only what the component\nreads. `secret` is opt-in: set `\"secret\": true` ONLY for credentials (they route\nthrough the secret path), and OMIT it entirely otherwise. Give each key a\n`description` saying what the value is and where the user finds it\n(`{ \"key\": \"STRIPE_API_KEY\", \"secret\": true, \"description\": \"Your Stripe secret\nAPI key\" }`) — the Build dependency drawer shows it under the field. For a\nNON-secret key whose sensible default you can infer, add `defaultValue` and the\ndrawer pre-fills it (`{ \"key\": \"AWS_REGION\", \"defaultValue\": \"us-east-1\" }`).\nNEVER set `defaultValue` on a secret — a credential has no default to invent. An\n`external` dependency should almost always carry at least one key: the\nvalue-collection gate needs something to collect.\n\n#### How the platform derives status\u002Freason\n\nYou never author `status`\u002F`reason`. The platform reads them off the dependency\nfile at read time, plus one registry lookup for a copy, first match wins:\n\n1. `resource.ref` set and the organization has a registered resource of that\n   name → the copy stands; only the contract is checked (rule 4)\n2. `resource.ref` set and NO registered resource of that name → `unresolved`\u002F\n   `needs-input` — the definition says \"The organization has no registered\n   resource with this name\" and offers Select a provider\n3. no `resource.provider` (`suggestions` open or not) → `unresolved`\u002F\n   `needs-input` — the user has not chosen a service; the definition asks them\n4. no `resource.contract`, or a contract whose file is not on disk (the\n   platform's fetch may still be owed or may have failed) → `unresolved`\u002F\n   `needs-contract`\n5. `contract.origin: \"assumed\"` with no `accepted` → `unresolved`\u002F\n   `needs-acceptance`\n6. otherwise → `resolved` — flagged `registered` for a copy, `assumed` when\n   the contract was accepted as an assumption, `derived` when it was written\n   from the provider's own documentation, `sdk-only` for an sdk contract,\n   `stale` when a copy's document no longer matches the organization's\n\n`component` is always `resolved` here. A `platform-resource` is too — once\nemitted — so only emit one whose `resourceType` is a `name` from this turn's\n`list_platform_resource_types`. An `org-service` resolves on catalog visibility,\nand is `blocked`\u002F`access-required` when the provider exists but this project\ncannot see it. `needsSpec`, `specPath` and the other definition fields are gone from the\ncomponent's schema — a draft carrying them fails the write-gate with a message\nnaming the dependency file they moved to.\n\n### Narrating the design turn\n\nThe design-generate turn runs in the chat panel, so your turn text is what the\nuser watches live. **Narrate each dependency decision in one plain-prose line as\nyou settle it** in the Dependencies step, before moving to the next:\n\n- resolved → `✓ \u003Ccapability>: using \u003Cchoice>` (say `, document fetched at\n  save` when you pointed at a published one, `, interface derived from docs`\n  when you wrote it from the provider's reference, `, registered — the\n  organization's resource, copied here` for a reuse)\n- needs-input → `\u003Ccapability>: your choice — A \u002F B \u002F C are common; select a\n  provider on its definition`\n- needs-contract → `\u003Ccapability>: \u003Cprovider> chosen, no published contract\n  found — you can upload one or let me assume it, from the dependency's definition in the spec view`\n\nNever block the design on an unresolved dependency — print the line\nand keep emitting the rest; the user replies in the same chat to steer it, now or\nlater. Then close as `design` says. Its **\"Needs your input\"** block lists ONLY\nthe dependencies still unresolved, each as a LINK to its definition —\n`[currency-service](aep:\u002F\u002Fspec\u002Fspecs\u002Fdesign\u002Fdependencies\u002Fcurrency-service\u002Fdependency.json)`\n— followed by the single thing you need (the console opens the definition\nfrom the link; this link form is the one place a repo path is allowed in\nyour prose).\nEach **Needs your input** line names the dependency the way its definition in the spec view does, so\nthe user can click through and press **Resolve** — that runs the\n`resolve-dependency` flow, which is where the contract gets provided or\nassumed; the design turn never waits for it.\n\n### Resolving or reconsidering a named dependency on request\n\n`\u002Fresolve-dependency \u003Cname>` runs the `resolve-dependency` skill — the guided\nflow that takes one dependency from open to resolved; read that skill when the\ninstruction names it. A plain chat turn may still point you at a dependency\n(\"reconsider `stripe`\"): read `specs\u002Fdesign\u002Fdependencies\u002F\u003Cname>\u002Fdependency.json`\nfrom the snapshot and act on its current state.\n\n- **No service chosen — it carries `suggestions`, or nothing.** The user's\n  answer, if the instruction carries one (a service name or a document URL),\n  IS the choice: set `provider` + `style`, remove `suggestions`, and go for\n  the contract (step 3). With no answer, put the choice to them — the\n  suggestions, or what your research finds, each with a one-line distinction,\n  plus that they may name another — and write nothing until they answer.\n- **Already resolved — reconsider.** This is the only branch that may leave a\n  catalog row that still fills the role. Present fresh alternatives in the\n  conversation, or repin to the Registered name the user picks, or start a\n  **Project External resource** under a **new** name.\n\nEdit ONLY that dependency's file: re-emit the whole `dependency.json` (never a\npatch) with every field you are not changing carried over exactly. The\ncomponents' references do not change. Never write or alter `assumed` — that is\nthe user's record; the write-gate refuses it.\n\n### Descriptions, and the per-component artifacts\n\nEvery dependency carries a one-line `description`: what the target is and how\nthis component uses it. Source it per kind — an `external`'s definition file\ncarries what the system is (the consumption instructions when the name is a\nRegistered External resource) and the component's reference why this component\ncalls it; an `org-service`'s\nsays the specific\noperations it calls from the discovered contract, or plainly that no contract was\nresolvable, never a guess; a `platform-resource`'s says what it stores. The\nconsole shows it in the dependency drawer and the coding agent relies on it to\nintegrate correctly.\n\nOne component per directory. Every `web-application` gets a prototype,\n`prototype.json` and `prototype.tsx` (`prototype` governs it); every `service`\ngets an `openapi.yaml` (`openapi-conventions` governs it); every `ai-agent`\ngets an `agent.afm.md` (`agent-building` governs it). `design` says when each\nis written. Other kinds (scheduled tasks, workers, …) carry no extra artifact\nyet — capture their behaviour fully in `description` and `dependencies`.\n",{"name":5,"metadata":24,"description":7},{"aep":25},{"kind":26,"audience":27},"platform",[28],"design","main","e3b5fc8668f26761380b2d4965719f2d8f2eb246","skills\u002Farchitecture","https:\u002F\u002Fgithub.com\u002Fwso2\u002Flabs-agentic-engineer","https:\u002F\u002Fgithub.com\u002Fwso2","https:\u002F\u002Fwso2.com","Welcome to the WSO2 source code! For info on working with the WSO2 repositories and contributing code, click the link below.","2026-11-11 15:36:27",{"source":8,"skillName":5,"commands":38,"prompt":48,"zipAvailable":49},[39,42,45],{"tool":40,"command":41},"npx","npx skills add wso2\u002Flabs-agentic-engineer --skill architecture",{"tool":43,"command":44},"bunx","bunx skills add wso2\u002Flabs-agentic-engineer --skill architecture",{"tool":46,"command":47},"pnpm","pnpm dlx skills add wso2\u002Flabs-agentic-engineer --skill architecture","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fwso2\u002Flabs-agentic-engineer\u002Fblob\u002Fmain\u002Fskills\u002Farchitecture\u002FSKILL.md",true,[],[52,59,66,73,80,87],{"skillNo":53,"name":54,"nameZh":6,"description":55,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":56,"categories":57,"keywords":58,"publishedAt":19},"01M735MP9NFFRPS260G9D16JQX","validation-task","Load when working a VALIDATION task dispatched by WSO2 Labs Agentic Engineer — judging a deployed version against its acceptance criteria.","https:\u002F\u002Fgithub.com\u002Fwso2\u002Flabs-agentic-engineer\u002Fblob\u002Fmain\u002Fskills\u002Fvalidation-task\u002FSKILL.md",[],[],{"skillNo":60,"name":61,"nameZh":6,"description":62,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":63,"categories":64,"keywords":65,"publishedAt":19},"01M735MJ8ZMRZR20A57TMF7R9Q","thunder-authentication","Apply when a component sits on the project's sign-in — a SPA whose users authenticate through the auth platform-resource dependency, or a protected backend those signed-in users call.","https:\u002F\u002Fgithub.com\u002Fwso2\u002Flabs-agentic-engineer\u002Fblob\u002Fmain\u002Fskills\u002Fthunder-authentication\u002FSKILL.md",[],[],{"skillNo":67,"name":68,"nameZh":6,"description":69,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":70,"categories":71,"keywords":72,"publishedAt":19},"01M735MFQMY8DY1EM7JKS784D3","task-planning","Use when planning a version's build from a design — the plan turn that writes the build's checklist with planBuild, one item per feature the version carries in build order, after a Foundation item when the version needs one.","https:\u002F\u002Fgithub.com\u002Fwso2\u002Flabs-agentic-engineer\u002Fblob\u002Fmain\u002Fskills\u002Ftask-planning\u002FSKILL.md",[],[],{"skillNo":74,"name":75,"nameZh":6,"description":76,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":77,"categories":78,"keywords":79,"publishedAt":19},"01M735MCJQ4EVH7G4ZFMC682Y4","start","Use when kicking off a project from its idea — the product pass that agrees the feature list and writes the product page, the product-wide page and a stub per feature — or re-running \u002Fstart on a project that already has requirements.","https:\u002F\u002Fgithub.com\u002Fwso2\u002Flabs-agentic-engineer\u002Fblob\u002Fmain\u002Fskills\u002Fstart\u002FSKILL.md",[],[],{"skillNo":81,"name":82,"nameZh":6,"description":83,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":84,"categories":85,"keywords":86,"publishedAt":19},"01M735M9R050KT3H34PVXN2MKK","security-design","Write specs\u002Fdesign\u002Fsecurity.json when a design has sign-in, permissions, roles or test users — the permission catalog every API operation is gated on.","https:\u002F\u002Fgithub.com\u002Fwso2\u002Flabs-agentic-engineer\u002Fblob\u002Fmain\u002Fskills\u002Fsecurity-design\u002FSKILL.md",[],[],{"skillNo":88,"name":89,"nameZh":6,"description":90,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":91,"categories":92,"keywords":93,"publishedAt":19},"01M735M60CTW4S7XMS6XCX9A7Y","resolve-dependency","Use for taking one external dependency from open to resolved — `\u002Fresolve-dependency \u003Cname> [answer]` names it and may carry the provider the user chose; `\u002Fresolve-dependencies` walks every open one in turn. Ask which provider (their suggestions as options; never choose for them), get its interface on disk (found, uploaded from the card, or assumed under the authorization the card records), then derive the config keys.","https:\u002F\u002Fgithub.com\u002Fwso2\u002Flabs-agentic-engineer\u002Fblob\u002Fmain\u002Fskills\u002Fresolve-dependency\u002FSKILL.md",[],[]]