@cat-factory/worker 0.6.0

package/migrations/0002_consensus.sql

@@ -0,0 +1,38 @@
+-- Consensus orchestration (optional `@cat-factory/consensus` mechanism) + the core
+-- task-estimator triage that gates it.
+--
+-- `blocks.estimate`     — the task-estimator's JSON triage (complexity/risk/impact),
+--                         persisted on the task block for gating + UI. CORE.
+-- `pipelines.consensus` — JSON array of per-step consensus configs, parallel to
+--                         agent_kinds (set in the pipeline builder for eligible steps).
+-- `consensus_sessions`  — one row per (execution, step) recording the multi-model
+--                         transcript (participants, rounds/votes, synthesis) — the
+--                         observability surface the dedicated window renders.
+
+ALTER TABLE blocks ADD COLUMN estimate TEXT;
+ALTER TABLE pipelines ADD COLUMN consensus TEXT;
+
+CREATE TABLE consensus_sessions (
+  workspace_id TEXT    NOT NULL,
+  id           TEXT    NOT NULL,
+  block_id     TEXT    NOT NULL,
+  execution_id TEXT,
+  step_index   INTEGER NOT NULL,
+  agent_kind   TEXT    NOT NULL,
+  strategy     TEXT    NOT NULL,
+  status       TEXT    NOT NULL,
+  participants TEXT    NOT NULL DEFAULT '[]',  -- JSON ConsensusParticipant[]
+  rounds       TEXT    NOT NULL DEFAULT '[]',  -- JSON ConsensusRound[]
+  synthesis    TEXT,                           -- the synthesized result (null until done)
+  confidence   REAL,                           -- aggregate confidence 0..1 (ranked-voting)
+  dissent      TEXT    NOT NULL DEFAULT '[]',  -- JSON string[]
+  error        TEXT,                           -- failure detail when status='failed'
+  created_at   INTEGER NOT NULL,
+  updated_at   INTEGER NOT NULL,
+  PRIMARY KEY (workspace_id, id)
+);
+
+CREATE INDEX idx_consensus_sessions_step
+  ON consensus_sessions (workspace_id, execution_id, step_index);
+CREATE INDEX idx_consensus_sessions_block
+  ON consensus_sessions (workspace_id, block_id, created_at);

package/migrations/0002_llm_reasoning_text.sql

@@ -0,0 +1,6 @@
+-- Capture the model's reasoning / "thinking" trace alongside the response text.
+-- A reasoning model (e.g. @cf/moonshotai/kimi-k2.7-code) can spend its whole output
+-- budget thinking and return an empty completion; without this column those output
+-- tokens were unaccounted for (response_text empty, no trace). Mirrored on Node by the
+-- Drizzle `reasoning_text` column.
+ALTER TABLE llm_call_metrics ADD COLUMN reasoning_text TEXT NOT NULL DEFAULT '';

package/migrations/0002_local_model_endpoints.sql

@@ -0,0 +1,16 @@
+-- Per-USER locally-run model endpoints (Ollama / LM Studio / llama.cpp / vLLM / custom
+-- OpenAI-compatible runners). A runner lives on the user's own machine, so endpoints are
+-- scoped to the user and resolved by the run initiator at execution time. The optional
+-- bearer key is stored system-key-encrypted (`api_key_cipher`); `models` is a JSON array
+-- of the model ids the user enabled from that runner.
+CREATE TABLE local_model_endpoints (
+  user_id        TEXT NOT NULL,
+  provider       TEXT NOT NULL,        -- runner type: ollama | lmstudio | llamacpp | vllm | custom
+  label          TEXT NOT NULL,
+  base_url       TEXT NOT NULL,
+  api_key_cipher TEXT,                 -- system-key ciphertext of an optional bearer key (NULL = keyless)
+  models         TEXT NOT NULL,        -- JSON array of enabled model ids
+  created_at     INTEGER NOT NULL,
+  updated_at     INTEGER NOT NULL,
+  PRIMARY KEY (user_id, provider)
+);

package/migrations/0003_pipeline_labels_archive.sql

@@ -0,0 +1,13 @@
+-- Pipeline builder improvements: per-step estimate gating + library organization.
+--
+-- `pipelines.gating`   — JSON array of per-step StepGating, parallel to agent_kinds.
+--                        An enabled entry makes the step run only when the task
+--                        estimate meets the threshold (else it is skipped at runtime).
+--                        Used today to make a companion (reviewer / architect-companion /
+--                        spec-companion) conditional on how heavy the task is.
+-- `pipelines.labels`   — JSON array of free-form organizational labels (filter/search).
+-- `pipelines.archived` — 1 when the pipeline is hidden from the default library view.
+
+ALTER TABLE pipelines ADD COLUMN gating TEXT;
+ALTER TABLE pipelines ADD COLUMN labels TEXT;
+ALTER TABLE pipelines ADD COLUMN archived INTEGER;

package/migrations/0003_release_health.sql

@@ -0,0 +1,34 @@
+-- Post-release-health gate (Datadog). Adds the two merge-preset knobs that bound the
+-- gate (how long it watches a deployed release's monitors/SLOs, and how many on-call
+-- investigations it may dispatch), and the two tables backing the integration: the
+-- per-workspace Datadog connection (credentials sealed at rest) and the per-block
+-- monitor/SLO mapping the gate reads. Mirrors the Drizzle schema in runtimes/node.
+
+ALTER TABLE merge_threshold_presets
+  ADD COLUMN release_watch_window_minutes INTEGER NOT NULL DEFAULT 30;
+ALTER TABLE merge_threshold_presets
+  ADD COLUMN release_max_attempts INTEGER NOT NULL DEFAULT 1;
+
+-- One Datadog connection per workspace. `api_key`/`app_key` are stored encrypted by the
+-- facade's SecretCipher (domain tag 'cat-factory:datadog'); plaintext only in memory.
+CREATE TABLE datadog_connections (
+  workspace_id TEXT    NOT NULL,
+  site         TEXT    NOT NULL,
+  api_key      TEXT    NOT NULL,
+  app_key      TEXT    NOT NULL,
+  created_at   INTEGER NOT NULL,
+  updated_at   INTEGER NOT NULL,
+  PRIMARY KEY (workspace_id)
+);
+
+-- Per-block (service frame) monitor/SLO mapping the post-release-health gate reads.
+CREATE TABLE release_health_configs (
+  workspace_id    TEXT    NOT NULL,
+  block_id        TEXT    NOT NULL,
+  monitor_ids     TEXT    NOT NULL DEFAULT '[]',
+  slo_ids         TEXT    NOT NULL DEFAULT '[]',
+  env_tag         TEXT,
+  created_at      INTEGER NOT NULL,
+  updated_at      INTEGER NOT NULL,
+  PRIMARY KEY (workspace_id, block_id)
+);

package/migrations/0004_run_timing_task_types.sql

@@ -0,0 +1,24 @@
+-- Run-timing rework + task types + per-service task limits.
+--
+-- 1. Tasks gain a type (feature/bug/document/spike/recurring) + small per-type form
+--    fields. Both are task-level; absent ⇒ treated as a 'feature' with no extra fields.
+ALTER TABLE blocks ADD COLUMN task_type TEXT;
+ALTER TABLE blocks ADD COLUMN task_type_fields TEXT;
+
+-- 2. Notifications gain a render severity. `normal` (yellow) until the escalation sweep
+--    flips a long-waiting one to `urgent` (red) — the signal that replaced the old
+--    hard "decision timeout" auto-fail. NULL is read as 'normal'.
+ALTER TABLE notifications ADD COLUMN severity TEXT;
+
+-- 3. Per-workspace runtime settings: the human-wait escalation threshold and the
+--    per-service running-task limit policy. One row per workspace; lazily seeded.
+CREATE TABLE workspace_settings (
+  workspace_id               TEXT    NOT NULL PRIMARY KEY,
+  waiting_escalation_minutes INTEGER NOT NULL DEFAULT 120,
+  -- 'off' | 'shared' | 'per_type'
+  task_limit_mode            TEXT    NOT NULL DEFAULT 'off',
+  -- The shared cap when task_limit_mode = 'shared'; NULL otherwise.
+  task_limit_shared          INTEGER,
+  -- JSON object of per-type caps when task_limit_mode = 'per_type'; NULL otherwise.
+  task_limit_per_type        TEXT
+);

package/package.json

@@ -0,0 +1,65 @@
+{
+  "name": "@cat-factory/worker",
+  "version": "0.6.0",
+  "description": "Reusable Cloudflare Worker library (Hono + D1) exposing the Agent Architecture Board core: the `createApp()` Hono factory, the default fetch/scheduled/queue handler, and the Durable Object + Workflow classes. Deployments (see deploy/backend) supply the wrangler.toml + config and re-export these.",
+  "files": [
+    "dist",
+    "migrations"
+  ],
+  "type": "module",
+  "main": "./dist/index.js",
+  "types": "./dist/index.d.ts",
+  "exports": {
+    ".": {
+      "types": "./dist/index.d.ts",
+      "default": "./dist/index.js"
+    },
+    "./app": {
+      "types": "./dist/app.d.ts",
+      "default": "./dist/app.js"
+    },
+    "./package.json": "./package.json"
+  },
+  "publishConfig": {
+    "access": "public"
+  },
+  "dependencies": {
+    "@ai-sdk/anthropic": "^3.0.85",
+    "@ai-sdk/openai": "^3.0.73",
+    "@ai-sdk/openai-compatible": "^2.0.51",
+    "@cloudflare/containers": "^0.3.7",
+    "@hono/valibot-validator": "^0.5.3",
+    "ai": "^6.0.208",
+    "hono": "^4.12.26",
+    "pino": "^9.14.0",
+    "valibot": "^1.4.1",
+    "workers-ai-provider": "^3.2.0",
+    "@cat-factory/agents": "0.6.0",
+    "@cat-factory/consensus": "0.6.0",
+    "@cat-factory/observability-langfuse": "0.6.0",
+    "@cat-factory/server": "0.6.0",
+    "@cat-factory/contracts": "0.6.0",
+    "@cat-factory/integrations": "0.6.0",
+    "@cat-factory/orchestration": "0.6.0",
+    "@cat-factory/kernel": "0.6.0",
+    "@cat-factory/spend": "0.6.0",
+    "@cat-factory/prompt-fragments": "0.6.0",
+    "@cat-factory/provider-cloudflare": "0.6.0"
+  },
+  "devDependencies": {
+    "@cloudflare/vitest-pool-workers": "^0.16.17",
+    "@cloudflare/workers-types": "^4.20260619.1",
+    "typescript": "7.0.1-rc",
+    "vitest": "^4.1.9",
+    "wrangler": "^4.102.0",
+    "@cat-factory/conformance": "0.6.0"
+  },
+  "scripts": {
+    "build": "tsc -b tsconfig.build.json",
+    "dev": "wrangler dev",
+    "typecheck": "tsc -p tsconfig.build.json --noEmit",
+    "test": "vitest",
+    "test:run": "vitest run",
+    "db:migrate:local": "wrangler d1 migrations apply cat_factory --local"
+  }
+}