alextiannus/iaic-core

Última atividade de origem registrada: 14 de setembro de 2026 às 15:46

Catálogo do SkillsMP atualizado: 9 de novembro de 2026 às 14:22

skills coletadas: 6

Estrelas no GitHub: 0
Forks no GitHub: 0

Instalar com um assistente de IA

Copie este prompt para o assistente de IA que você usa.

Follow https://skillsmp.com/skill-install/prompt.md to install Agent Skills from https://github.com/alextiannus/iaic-core.

README do repositório

Arquivo de origem do README: README.md

Ler o README original

Expandir README completo

# IAiC Core Catalog validation fix (2026-09-19 UTC): candidate.108 rejects sparse catalog arrays instead of incorrectly passing an empty expectation. Sparse matrix configuration is rejected before any reader runs. No authorization/runtime changes. Access matrix update (2026-09-18): candidate.107 adds an optional typed Host catalog matrix checker for explicit Actor/context fixtures across HTTP/MCP/UI entrances and model/host surfaces. It reports omissions and excess visibility without invoking capabilities. See developer/ACCESS_MATRIX.md. Execution matrix automation and application acceptance remain pending. Access diagnostics update (2026-09-18): candidate.106 adds stable public codes for unavailable capabilities, Domain denial, Task scope exclusion and missing Actor. HTTP status and MCP protocol/tool-error envelopes are preserved. MCP no longer exports arbitrary private error.code values. See capabilities/ACCESS_DIAGNOSTICS.md. Actor matrix tooling and automatic projection drift detection remain pending. Model readiness update (2026-09-18): candidate.105 adds an opt-in typed evidence gate for Host-declared job requirements, bound verification, expiry/revocation and HTTPS/TLS evidence. Runtime checks admission; metering checks before reserve; the provider wrapper rechecks before dispatch. See agent/MODEL_READINESS.md. Real probe execution, multimodal transport and first-class Task requirements are not included. Existing unknown usage is never released by this gate. Capability visibility update (2026-09-18): candidate.104 adds opt-in model/host/both consumption declarations. Model tools, HTTP/MCP discovery and execution, Runtime calls and historical call projections enforce the selected surface. Defaults keep undeclared capabilities on both surfaces. Host-only still requires current Actor permission, execution policy and stable write keys. See capabilities/VISIBILITY.md for explicit Host composition, migration, old Task handling and remaining limits. MCP type update (2026-09-18): candidate.103 provides the typed server subpath, peer SDK AuthInfo access context and an optional HTTP connection helper. A real installed HTTP consumer checks current identity, permission and stable write keys. SDK 1.30.0 still has two upstream exact-optional declaration errors; these are explicitly checked, not reported as a clean full-library compile. See mcp/README.md. OAuth provider implementation and full authentication acceptance remain Host work. HTTP result update (2026-09-17): candidate.102 adds opt-in typed `invokeResult` for deterministic business results; `invoke` retains its transport envelope. Malformed responses retain observed HTTP status and never trigger automatic retry. See http/README.md for migration, unknown outcomes and the separate pending visibility/Artifact-transfer work. Application adoption remains unverified. Typed starter update (2026-09-17): candidate.101 supplies the ordinary Agent composition's strict NodeNext declaration and an independently installed PostgreSQL consumer. Task admission, wait/rebuild/continuation, explicit Host Actor restoration and allowance settlement are exercised. See developer/templates/agent/README.md for the exact scope: raw module handles remain unknown; optional extension and full root/Agent/Task/Peer declarations are not complete. No runtime/schema change. Host context update (2026-09-17): candidate.100 adds Host-only immutable context references, separate model projections, explicit current business Actor restoration, and ordinary Agent starter wiring. See [Host context contract](context/HOST.md). Full starter/root Task declarations remain pending; application schemas, redaction, authority and production acceptance remain with each application. Executor handoff update (2026-09-17): candidate.99 adds trusted generation-scoped drain requests, fenced executor transactions and a bounded Runtime drain port. In-flight work may finish; the old Runtime retires before another Task claim. See [deployment contract](agent/EXECUTOR_HANDOFF.md). First upgrades from older binaries require a controlled executor stop; real rollout acceptance stays with the application. Host Context/template requests remain pending. Model failure update (2026-09-17): candidate.98 distinguishes measured output truncation, invalid actions, provider failures and model deadlines. Deterministic failures cannot be blindly resumed; unknown usage still requires reconciliation. See [failure and migration contract](agent/MODEL_FAILURES.md). Executor deployment handoff and the agent-template typing request remain pending. Lark type fix (2026-09-15): candidate.97 supplies declarations for both Lark channel exports, verified by an independent strict NodeNext consumer and official SDK structural compatibility. Runtime behavior is unchanged; full root and Agent/Task/Peer types remain pending. See channels/LARK.md. Provider integration update (2026-09-15): candidate.96 adds WhatsApp Cloud API and WeCom internal-application text channels, plus Telegram outbound delivery and verified webhook ingress. All share the existing User Assistant and Outbox. The WeCom optional encrypted entry uses open-source crypto/XML packages; WhatsApp requires a current host text-window policy. Live account validation, attachments/cards and provider Unknown recovery remain pending. See [provider integration](channels/PROVIDERS.md). Lark integration update (2026-09-15): candidate.95 adds inbound text and SDK outbound delivery, plus an optional official Lark MCP bridge with 89 reviewed tools in eight domains, including group join/member management. Real official MCP schemas and handlers are exercised with a substituted SDK network boundary; IM/Task/Outbox integration uses PostgreSQL. Live tenant credentials, subscriptions and application acceptance remain pending. See [Lark integration](channels/LARK.md). TypeScript scope update (2026-09-15): candidate.94 adds declarations for the Capability/Dispatcher and HTTP server/client module entries, plus structural error and Task admission types. A separate tarball consumer compiles under strict NodeNext and executes its emitted program. Root and full Agent/Task/Peer module coverage remain pending; see capabilities/TYPES.md for the exact supported imports. Package identity update (2026-09-15): candidate.93 uses internal npm version `0.1.0-candidate.93` and its version-qualified tarball filename. `CORE_RELEASE` reports the installed package identity. The release gate includes an ordinary candidate.92 → current npm upgrade with the same cache, lockfile and node_modules. See releases/IDENTITY.md. This does not complete the separate public TypeScript surface request or prove application integration. Current scope addition (2026-09-15): IM is a replaceable User Assistant interface. The channels module adds a durable inbox, explicit cross-channel account binding, Core task controls and notification projection, with Telegram/Slack text normalizers. The local file module adds authorized UTF-8 read/create and original receipts; existing browser tools cover host-bound form entry. See channels/README.md and examples/core-assistant-channels. Real IM authentication/delivery installations, a remote device transport and general desktop control remain application work. A headless JavaScript framework for AI Native Applications: configure an Agent's responsibility and resources, give it a goal, let it use authorized capabilities, and retain work and results beyond a chat connection. This is the **0.1.0 public source candidate**, published at https://github.com/alextiannus/iaic-core. A reuse license has not yet been selected; public visibility does not imply an MIT or other open-source license grant. See [ACCEPTANCE.md](ACCEPTANCE.md) for the current requirement-to-evidence map and [STATUS.md](STATUS.md) for the chronological implementation record. ## Start here Read the [AI Native Application](AI_NATIVE_APPLICATION.md) definition before choosing Agent roles or composing capabilities. It is the canonical conceptual contract for represented responsibility, authority, deterministic Harness boundaries, evidence, and accountable evolution. Application developers and Core contributors submit reusable Framework findings through the [Framework feedback guide](FRAMEWORK_FEEDBACK.md). Runtime and private reports remain in the application's Support lifecycle until an authorized developer has redacted them and established a reusable Core concern. Requirements currently verified by this candidate: Node.js 20, npm, and an isolated PostgreSQL database for persistent examples. PostgreSQL connections belong to the host application; Core accepts an injected pool. ```sh npm ci export SUBMISSION_TEST_DATABASE_URL='postgresql://localhost/iaic_example' npm run example:jobs ``` Use your own database connection. The examples create temporary synthetic schemas and remove them when finished. Default examples use deterministic model fixtures, not a paid model endpoint. The configured-jobs example combines user-plus-organization scopes, an organization-shared job, Memory, Workspace, sourced Knowledge, selected Skills, models and platform allowance, Session continuity, model-directed scheduling and an external MCP client. It includes application policy to illustrate composition; it does not prescribe job types or business flows. ```sh docker pull node@sha256:8094c002d08262dba12645a3b4a15cd6cd627d30bc782f53229a2ec13ee22a00 npm test # module checks and standalone examples npm run verify:core-package # pack, install elsewhere, run examples there npm pack # create a reusable Core npm archive ``` The full test suite and package verification also require a running Docker daemon and the pinned image above for actual sandbox execution. An application installs the resulting `.tgz` and imports `@immedi/iaic-core` or a module subpath. It does not import a platform server or copy Core into every application. See [GETTING_STARTED.md](GETTING_STARTED.md). Independent-principal communication is available as the optional [Peer collaboration module](collaboration/peer/README.md), with a [runnable composition](examples/core-peer-collaboration/README.md). The optional [User AI Assistant demo](examples/core-user-assistant/README.md) shows private resources, clarification and user-authorized declaration submission, with a system-model server and a minimal terminal client. ## Basic modules | Need | Modules and contracts | | --- | --- | | Simulate local Agent work | [Local simulation](developer/SIMULATION.md): scripted models, scoped tools and original response-loss receipts | | Develop and migrate an application | [Developer tools](developer/README.md): CLI, independent starter and versioned PostgreSQL migrations | | Manage accounts and organizations | [Accounts](accounts/README.md): replaceable directory, memberships and current identity mapping | | Record money and execute payments | [Payments](payments/README.md): confirmed invoice sources, charge/refund intents, provider adapters and reconciliation | | Resolve subscription access | [Subscriptions](subscriptions/README.md): versioned plan snapshots, current periods and entitlements | | Deliver background results | [Notifications](notifications/README.md): durable queue, channel ports, attempts and reconciliation | | Evaluate capability changes | [Evaluation](evaluation/README.md): independent grading, complete case coverage, baselines and evidence | | Observe running releases | [Observation](observation/README.md): trusted metrics, fixed policies and atomic protective rollback | | Select evaluated releases | [Releases](releases/README.md): immutable version bindings, canary selection, stop and rollback | | Store binary resources | [Object storage](storage/README.md) and [release resource loading](releases/README.md): scoped content references and verified files | | Observe and control a browser device | [Devices](devices/README.md): host-owned Page adapter, current policy and persistent action receipts | | Execute released code | [Execution](execution/README.md): pinned Docker sandbox, resource loading and shared code capability | | Describe executable operations | [Capabilities](capabilities/README.md): schemas, authorization, functions or Agent implementations, result verification | | Persistent work and model loop | [Agent](agent/README.md), [Tasks](tasks/README.md), [Context](context/README.md) | | Configure jobs and long-lived work identity | [Identities](identities/README.md): `AgentRegistry.register`, host JSON configuration, revision-bound tasks | | Remember, reference and produce | [Memory](memory/README.md), [Knowledge](knowledge/README.md), [Workspace](workspace/README.md) | | Discover installed working methods | [Skills](skills/README.md): selected catalogs and progressive reads | | Route and bound model requests | [Explicit routing](assistants/ROUTING.md), [shared concurrency](agent/CAPACITY.md), [request/Token rates](agent/RATES.md), [currency estimates](costs/README.md) | | Clean expired current resources | [Retention](resources/RETENTION.md): opt-in bounded Memory/Knowledge cleanup and version tombstones | | Choose a model and account for use | [Assistants](assistants/README.md), [Credentials](credentials/README.md), [Billing](billing/README.md) | | Continue conversations and proactive work | [Sessions](sessions/README.md), [Deferred](deferred/README.md), [Events](events/README.md), [Triggers](triggers/README.md), [Recurring](recurring/README.md) | | Authorize and cooperate | [Mandates](mandates/README.md), [Handoffs](handoffs/README.md) | | Connect an external Agent | [A2A](a2a/README.md), [MCP](mcp/README.md), [HTTP API and remote SDK](http/README.md), ordinary ESM interfaces | Each module owns its state and documents its injected dependencies. [Assistant work composition](assistants/TASKS.md) combines these ports without another Runtime or container framework. ## Application boundary Core provides basic capabilities, interfaces and necessary runtime rules. Applications own authenticated identity mapping, business truth and calculations, capability implementations, installed Skill code, configuration storage/distribution, resource-sharing policy, service deployment and UI/UX. Core prioritizes **Platform AI capabilities**, implemented by a team of a built-in Agent driven by the system model and the current external Codex Agent. Shared tasks, work artifacts and feedback support mutual takeover and reciprocal review; they are complementary participants in the same role. Codex need not adopt Core Runtime. The team still needs platform-side capabilities for development, evaluation and release; the existing CLI and evidence tools alone do not complete this integration. **User Assistant AI foundations** remain the next priority. User Assistants may be application-provided or external third-party Agents. Business AI is application-resident, using system model configuration and system/server capabilities to fulfill the application's business responsibilities. Core only supplies cross-application Business AI capabilities where useful; a built-in Business AI product is not required. Task length and complexity do not determine responsibility. See [design and delivery requirements](CORE_REQUIREMENTS.md). Default long-lived resource scoping should distinguish user plus organization. Applications can explicitly share company resources or configure a shared job. A shared resource scope does not automatically grant access to every member's Task history. A personal external Agent can call authorized capabilities without adopting Core Runtime. ## Source and validation [SOURCE.json](SOURCE.json) records the source revision and hashes of extracted files. The candidate includes Core modules and neutral examples only; private application implementations, credentials, production data and private Git history are not included. Some inherited module notes refer to historical integration locations; those are context, not runtime dependencies or files required by this candidate. Core-only module checks now live in `test/` and run with `npm run test:modules`; application-coupled integration suites remain in the application repository. Each check imports the Core package directly. The independent CI runs module checks and the standalone examples from a fresh package installation. These checks establish the documented composition contracts; they do not establish universal model quality or completion of every AI Native Application requirement. The complete delivery target and acceptance conditions are tracked in [CORE_REQUIREMENTS.md](CORE_REQUIREMENTS.md). Current usability does not imply that all conditions are satisfied. Every Platform AI Team member maintains system documentation for its changes and reviews its peer's documentation alongside code. Superseded versions have finite [version lifetimes and migration instructions](VERSION_LIFECYCLE.md). A User Assistant AI Demo is not a Core deliverable or acceptance requirement. A minimal [Platform Team host](examples/core-platform-team/README.md) assembles the existing modules with a system model, separate development allowance, shared work and reciprocal reviews. It accepts work requests from either member and retains requester/executor provenance. A real system-model/Codex review, correction and reciprocal-review loop has also completed; see the host evidence below. Arbitrary external takeover of an interrupted native Task remains unfinished. ### User issue resolution The additive [Support issue lifecycle](support/README.md) connects authenticated user feedback, engineering status, verified release evidence and notification admission. It is a Basic framework mechanism with application-owned capture, permissions, release verification and notification delivery adapters. The current support candidate is local and unpublished.

Carregando o README do repositório…

Skills neste repositório

classificação pendente

skill

ocupação

descrição

atualizado

declarations

sem classificação

Help the current user prepare, clarify, submit and verify a declaration through the host application.