refactor: repo-wide deslop (net −51.8k lines)

.dockerignore

@@ -17,7 +17,4 @@
 *.log
 /*.png
 **/*.tsbuildinfo
-coverage
 node_modules
-qa-output
-supabase/.temp

.editorconfig

@@ -11,11 +11,5 @@ indent_size = 2
 [*.py]
 indent_size = 4
 
-[*.toml]
-indent_size = 4
-
 [*.md]
 trim_trailing_whitespace = false
-
-[Makefile]
-indent_style = tab

.gitignore

@@ -6,11 +6,7 @@ node_modules/
 .next/
 .vercel/
 dist/
-coverage/
 .qa/
-qa-output/
-qa-*.png
-*.har
 *.tsbuildinfo
 .env
 .env.*

.vercelignore

@@ -1,7 +1,3 @@
-# Local documentation and planning directories are not part of the deploy bundle.
-/docs/
-/tasks/
-/plans/
 # Build artifacts (match anywhere; Vercel installs + builds fresh)
 node_modules
 .pnpm-store
@@ -17,4 +13,3 @@ node_modules
 compose.yaml
 .dockerignore
 infra/containers/dev
-supabase

AGENTS.md

@@ -39,8 +39,8 @@ apps/                    Deployable services
   webhooks-worker/       Polar/Clerk/Composio webhooks + internal ops workflows
 
 packages/                Shared libraries
-  agent-core/            Mastra instance + workflows
-  tools-*/               Tool implementations per domain
+  agent-core/            Mastra instance, workflows, and data/document/media tools
+  tools-{browser,code,research}/  Multi-consumer and isolated tool-domain packages
   db/                    Drizzle schema (per-domain) + queries + migrations
   byok/                  Vault-backed BYOK key store
   skills/                Build-time skill bundler
@@ -49,9 +49,9 @@ packages/                Shared libraries
   types/                 Zod schemas + branded IDs + InferAgentUIMessage
   ui/                    Shared Cheatcode UI primitives, icon barrel, AI response renderer
 
-skills/                  8 curated Anthropic SKILL.md skills
-infra/                   Wrangler configs, Supabase migrations, Daytona sandbox Dockerfile
-scripts/                 Operational helpers only: build skills, local startup, deploy orchestration, migrations
+skills/                  19 curated skills; _shared/office is vendored and not bundled
+infra/                   Container images and Dockerfiles for local development and Daytona sandboxes
+scripts/                 Operational helpers: build-skills, dev, dev-worker-config, jsonc, migrate, migration-drizzle, database-operation-safety
 ```
 
 ## Build
@@ -73,46 +73,24 @@ pnpm turbo lint                         # Biome check (fails CI on warnings)
 pnpm turbo build                        # Production build
 ```
 
-Product/acceptance testing is direct `agent-browser --auto-connect --session cheatcode-debug`
-UI interaction only: click/fill/type through real flows, capture screenshots,
-inspect console/network output, and read app logs. Do not add or run scripted
-test harnesses for product flows, browser automation, prompt submission, auth,
-accessibility, load, or final E2E. Package `test` scripts are intentionally not
-part of the V2 command surface, and source-level `*.test.ts` files are
-intentionally absent. Do not generate temporary validation scripts either;
-operate the UI directly and check logs, and remove any throwaway product QA
-script that appears in the V2 tree. Operational scripts may exist only for
-build, migration, and local-stack configuration; they are not
-product tests and must not simulate UI/user flows. Do not create temporary
-testing scripts in `scripts/`, package folders, `/tmp`, or any out-of-tree
-location.
-
-May 27, 2026 user override: never use scripts for product testing. Product QA
-means direct `agent-browser` UI operation, screenshots, console/network
-inspection, and running app-log review only. Delete any future V2 product-flow
-validator, prompt runner, browser wrapper, or throwaway QA helper on sight
-instead of running it. The removed V1 tree must not be restored, copied back,
-or used as a source of product-test scripts.
-
-May 28, 2026 hardening: do not wrap product QA in `pnpm`, `tsx`, shell loops,
-`/tmp` helpers, generated files, browser-driver wrappers, package aliases, or
-any scripted flow. Every product UI action, screenshot, console read,
-network/resource inspection, and app-log inspection must be issued directly in
-the transcript. Typecheck/lint/build are code-health gates only; they are not
-product QA.
-
-May 28, 2026 direct override: delete any product-flow test script, temporary
-helper, command-loop runner, browser wrapper, prompt driver, curl flow, or
-package alias when discovered. The active V2 tree should contain no product-test
-scripts. The remaining `scripts/` files are operational only and must not click
-the UI, submit prompts, drive auth, gather acceptance evidence, or replace
-direct `agent-browser` operation.
-
-May 28, 2026 latest user directive: code the all-weeks V2 surface first, then
-run final product QA only through direct `agent-browser --auto-connect --session
-cheatcode-debug` UI actions and direct console/network/app-log inspection. Do
-not write, run, or keep scripts to submit prompts, click UI, drive auth, wrap
-`agent-browser`, run curl flows, or gather acceptance evidence.
+Product/acceptance testing uses only direct
+`agent-browser --auto-connect --session cheatcode-debug` commands issued in the
+transcript to click, fill, and type through real flows, capture screenshots,
+inspect console/network/resources, and read running app logs. Never add, run,
+write, or keep scripts for product testing, browser automation, prompt
+submission, auth, accessibility, load, or final E2E, including `pnpm`/`tsx`
+wrappers, shell or command loops, `/tmp` or generated helpers, browser-driver
+wrappers, package aliases, prompt drivers, curl flows, validators, and
+throwaway QA helpers in `scripts/`, package folders, or any out-of-tree
+location; delete them on sight instead of running them. Package `test` scripts
+and source-level `*.test.ts` files remain absent. Operational scripts may exist
+only for build, migration, and local-stack configuration and must not click the
+UI, submit prompts, drive auth, gather acceptance evidence, or replace direct
+browser operation. Typecheck, lint, and build are code-health gates only, not
+product QA. Code the all-weeks V2 surface first, then perform final QA through
+the direct browser commands and direct console/network/app-log inspection. The
+removed V1 tree must not be restored, copied back, used as a testing surface, or
+used as a source of product-test scripts.
 
 ## Run locally
 
@@ -121,47 +99,8 @@ pnpm dev                                # Compose: Next + chained Workers agains
 pnpm dev:down                           # Stop the local Compose stack
 ```
 
-Required local env vars in `.env.local` (template in `.env.example`):
-
-```
-# Production Supabase session-pooler URLs for the three isolated Worker roles.
-# Administrative migration credentials never belong in .env.local.
-SUPABASE_GATEWAY_DATABASE_URL=
-SUPABASE_AGENT_DATABASE_URL=
-SUPABASE_WEBHOOKS_DATABASE_URL=
-
-# Per-Worker signed tenant context (three distinct secrets, each at least 32 bytes)
-DATABASE_CONTEXT_SIGNING_SECRET_GATEWAY=
-DATABASE_CONTEXT_SIGNING_SECRET_AGENT=
-DATABASE_CONTEXT_SIGNING_SECRET_WEBHOOKS=
-
-# Daytona
-DAYTONA_API_KEY=
-DAYTONA_API_URL=https://app.daytona.io/api
-DAYTONA_TARGET=us
-DAYTONA_SANDBOX_SNAPSHOT=
-DAYTONA_WORKSPACE_VOLUME=cheatcode-workspaces-development
-PREVIEW_TOKEN_SECRET=
-
-# Clerk development instance only
-NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY=
-CLERK_SECRET_KEY=
-
-# Polar
-POLAR_ACCESS_TOKEN=
-POLAR_WEBHOOK_SECRET=
-
-# Composio
-COMPOSIO_API_KEY=
-COMPOSIO_AUTH_CONFIGS={"github":"ac_...","gmail":"ac_...","slack":"ac_...","notion":"ac_...","linear":"ac_..."}
-COMPOSIO_WEBHOOK_SECRET=
-
-# Signed output capability
-OUTPUT_DOWNLOAD_SIGNING_SECRET=
-
-# Browser-visible local routing configuration (not secrets)
-NEXT_PUBLIC_GATEWAY_URL=http://127.0.0.1:8787
-```
+Copy the complete `.env.local` template from `.env.example`;
+`scripts/dev.ts` validates it before local startup.
 
 Never commit `.env.local`. It is the sole laptop application credential file;
 its database URLs contain only the three least-privilege production runtime

CLAUDE.md

@@ -37,11 +37,9 @@ apps/
   webhooks-worker/        Clerk, Polar, Composio webhooks + internal ops workflows
 
 packages/
-  agent-core/             Mastra agent, workflows, tools, and runtime contexts
+  agent-core/             Mastra agent, workflows, contexts, and data/document/media tools
   tools-code/             Sandbox shell/file/git/runCode tools
   tools-browser/          Stagehand LOCAL browser automation
-  tools-docs/             pptxgenjs, docx, exceljs, @react-pdf/renderer
-  tools-data/             Arquero CSV analysis + deterministic SVG charts
   tools-research/         Exa + Firecrawl
   db/                     Drizzle schema (per-domain) + queries + migrations
   byok/                   Vault-backed BYOK with provider validation
@@ -54,9 +52,9 @@ packages/
   ui/                     shared UI primitives, icon barrel, AI response renderer
   tsconfig/               Shared base/nextjs/worker/library configs
 
-skills/                   8 curated Anthropic SKILL.md skills
-infra/                    Wrangler configs, Supabase migrations, Daytona sandbox Dockerfile/snapshot
-scripts/                  Operational helpers only: build skills, local startup, deploy orchestration, migrations, audit archive
+skills/                   19 curated skills; _shared/office is vendored and not bundled
+infra/                    Container images and Dockerfiles for local development and Daytona sandboxes
+scripts/                  Operational helpers: build-skills, dev, dev-worker-config, jsonc, migrate, migration-drizzle, database-operation-safety
 ```
 
 ## Critical conventions (non-negotiable)
@@ -89,42 +87,29 @@ pnpm --filter @cheatcode/db db:generate  # Create new migration
 ```
 
 Pre-commit (Lefthook) runs Biome on staged + typecheck on changed packages. Must stay <5s.
-Product/acceptance testing is direct `agent-browser --auto-connect --session cheatcode-debug`
-UI operation plus console/network/app-log review. Do not add or run scripted
-browser/product-flow test harnesses; package `test` scripts and source-level
-`*.test.ts` files are intentionally absent from the V2 command surface. Do not generate temporary validation scripts either; operate
-the UI directly and check logs, and remove any throwaway product QA script that
-appears in the V2 tree. Operational scripts may exist only for build,
-migration, secret sync, Docker cleanup, and guarded deploy orchestration; they are not
-product tests and must not simulate UI/user flows. Do not create temporary
-testing scripts in `scripts/`, package folders, `/tmp`, or any out-of-tree
-location. Delete future V2 product validators instead of running them. The
+Product/acceptance testing uses only direct
+`agent-browser --auto-connect --session cheatcode-debug` commands issued in the
+transcript to click, fill, and type through real flows, capture screenshots,
+inspect console/network/resources, and read running app logs. Never add, run,
+write, or keep scripts for browser or product-flow testing, including
+`pnpm`/`tsx` wrappers, shell or command loops, `/tmp` or generated helpers,
+browser-driver wrappers, package aliases, prompt drivers, curl flows,
+validators, and throwaway QA helpers in `scripts/`, package folders, or any
+out-of-tree location; delete them on sight instead of running them. Package
+`test` scripts and source-level `*.test.ts` files remain absent. Operational
+scripts may exist only for build, migration, secret sync, Docker cleanup, and
+guarded deploy orchestration and must not click the UI, submit prompts, drive
+auth, gather acceptance evidence, simulate UI/user flows, or replace direct
+browser operation. Typecheck, lint, and build are code-health gates only, not
+product QA. Code the all-weeks V2 surface first, then perform final QA through
+the direct browser commands and direct console/network/app-log inspection. The
 removed V1 tree must not be restored or copied back as a testing surface.
 
-May 28, 2026 hardening: do not wrap product QA in `pnpm`, `tsx`, shell loops,
-`/tmp` helpers, generated files, browser-driver wrappers, package aliases, or
-any scripted flow. Each UI action, screenshot, console read, network/resource
-inspection, and app-log inspection must be issued directly in the transcript.
-Typecheck/lint/build remain code-health gates only.
-
-May 28, 2026 direct override: delete any product-flow test script, temporary
-helper, command-loop runner, browser wrapper, prompt driver, curl flow, or
-package alias when discovered. The active V2 tree should contain no product-test
-scripts. The remaining `scripts/` files are operational only and must not click
-the UI, submit prompts, drive auth, gather acceptance evidence, or replace
-direct `agent-browser` operation.
-
-May 28, 2026 latest user directive: code the all-weeks V2 surface first, then
-run final product QA only through direct `agent-browser --auto-connect --session
-cheatcode-debug` UI actions and direct console/network/app-log inspection. Do
-not write, run, or keep scripts to submit prompts, click UI, drive auth, wrap
-`agent-browser`, run curl flows, or gather acceptance evidence.
-
 ## Where things live
 
 | Need | File |
 |---|---|
-| Add a new tool | `packages/tools-<domain>/src/<tool>.ts` |
+| Add a new tool | `packages/agent-core/src/tools/<domain>/<tool>.ts` for data/docs/media; `packages/tools-<domain>/src/<tool>.ts` for browser/code/research |
 | Add a new agent | `packages/agent-core/src/mastra/agents/<name>.ts` |
 | Add a new workflow | `packages/agent-core/src/mastra/workflows/<name>.ts` |
 | Add a new skill | `skills/<name>/SKILL.md` (+ optional `references/` / `assets/`) |
@@ -136,7 +121,13 @@ not write, run, or keep scripts to submit prompts, click UI, drive auth, wrap
 
 Curated skills are bundled at build time into `packages/skills/src/generated.ts` (Workers have no filesystem at runtime). Anthropic SKILL.md format. V2 has no bundled skill scripts, no `evals/evals.json`, no local skill-eval runner, and no `skill_run_script` tool.
 
-The source-of-truth catalog is the set of skill folders under `skills/`; do not duplicate a manually maintained name list here. External skill registry exports, skills.sh links, public publishing scripts, and launch-prep copy are outside V2 unless the user explicitly re-expands the plan.
+The source-of-truth catalog is the set of non-underscore skill folders under
+`skills/`; do not duplicate a manually maintained name list here.
+`skills/_shared/office/` is the shared vendored tree materialized by the sandbox
+Dockerfile, and `scripts/build-skills.ts` skips `_`-prefixed directories.
+External skill registry exports, skills.sh links, public publishing scripts,
+and launch-prep copy are outside V2 unless the user explicitly re-expands the
+plan.
 
 The bundler contract lives in `scripts/build-skills.ts` and `packages/skills`.
 

apps/agent-worker/README.md

@@ -194,9 +194,6 @@ Mastra tool-call chunks also emit `step_started`, `step_completed`,
 live stream. If the last stream subscriber disconnects while a run is still
 running, AgentRun emits `run_abandoned` for the watchdog/funnel trail.
 
-AgentRun also emits `data-plan` and `data-task-status` UI chunks so the web app can
-render task progress without polling a separate status endpoint.
-
 Project deletion first fences project/thread mutations, refuses an active run, records a
 durable cleanup request, then removes that project's workspace folder. The database marks
 cleanup complete only after the Agent service succeeds, so a repeated DELETE retries a

apps/agent-worker/package.json

@@ -7,7 +7,6 @@
   "types": "./dist/index.d.ts",
   "scripts": {
     "build": "wrangler deploy --dry-run",
-    "deploy": "wrangler deploy",
     "lint": "biome check .",
     "typecheck": "tsc -p tsconfig.json --noEmit"
   },

apps/agent-worker/src/agent-api-run-routes.ts

@@ -27,6 +27,7 @@ import {
   type AgentRunAdmissionOutcome,
   activeRunForThreadRoute,
   agentRunForRunId,
+  callAgentRun,
   fetchAgentRun,
   reconcileAgentRunAdmission,
   runEntitlementPolicy,
@@ -63,7 +64,6 @@ type RejectedRunResult = Exclude<
 export function registerAgentRunHttpRoutes(app: Hono<{ Bindings: AgentEnv }>): void {
   app.post("/v1/threads/:threadId/runs", createRun);
   app.get("/v1/threads/:threadId/runs/stream", streamActiveRun);
-  app.get("/v1/threads/:threadId/runs/status", activeRunStatus);
   app.post("/v1/runs/:runId/cancel", cancelRun);
   app.get("/v1/threads/:threadId/browser-takeover", browserTakeoverStatus);
   app.post("/v1/threads/:threadId/browser-takeover/start", startBrowserTakeover);
@@ -276,26 +276,11 @@ async function streamActiveRun(c: AgentContext): Promise<Response> {
   );
 }
 
-async function activeRunStatus(c: AgentContext): Promise<Response> {
-  const userId = readGatewayUserId(c.req.raw.headers);
-  const threadId = parseThreadRouteParam(c.req.param("threadId") ?? "");
-  const run = await activeRunForThreadRoute(c.env, userId, threadId);
-  if (!run) {
-    return new Response(null, { status: 204 });
-  }
-  return fetchAgentRun(agentRunForRunId(c.env, run.runId), "https://agent-run.internal/status", {
-    headers: { "X-Cheatcode-User-Id": userId },
-  });
-}
-
 async function cancelRun(c: AgentContext): Promise<Response> {
   const userId = readGatewayUserId(c.req.raw.headers);
   const runId = parseRunRouteParam(c.req.param("runId") ?? "");
   const run = await runForRoute(c.env, userId, runId);
-  return fetchAgentRun(agentRunForRunId(c.env, run.runId), "https://agent-run.internal/cancel", {
-    headers: { "X-Cheatcode-User-Id": userId },
-    method: "POST",
-  });
+  return callAgentRun(agentRunForRunId(c.env, run.runId).cancel(userId));
 }
 
 async function browserTakeoverStatus(c: AgentContext): Promise<Response> {
@@ -304,11 +289,7 @@ async function browserTakeoverStatus(c: AgentContext): Promise<Response> {
   if (!run) {
     return Response.json(BrowserTakeoverStatusSchema.parse({ status: "inactive" }));
   }
-  return fetchAgentRun(
-    agentRunForRunId(c.env, run.runId),
-    "https://agent-run.internal/browser-takeover",
-    { headers: { "X-Cheatcode-User-Id": userId } },
-  );
+  return callAgentRun(agentRunForRunId(c.env, run.runId).browserTakeoverStatus(userId));
 }
 
 async function startBrowserTakeover(c: AgentContext): Promise<Response> {
@@ -320,11 +301,7 @@ async function startBrowserTakeover(c: AgentContext): Promise<Response> {
       retriable: false,
     });
   }
-  return fetchAgentRun(
-    agentRunForRunId(c.env, run.runId),
-    "https://agent-run.internal/browser-takeover/start",
-    { headers: { "X-Cheatcode-User-Id": userId }, method: "POST" },
-  );
+  return callAgentRun(agentRunForRunId(c.env, run.runId).browserTakeoverStart(userId));
 }
 
 async function resumeBrowserTakeover(c: AgentContext): Promise<Response> {
@@ -336,17 +313,8 @@ async function resumeBrowserTakeover(c: AgentContext): Promise<Response> {
   if (!run) {
     return Response.json(BrowserTakeoverResumeResultSchema.parse({ ok: true, status: "inactive" }));
   }
-  return fetchAgentRun(
-    agentRunForRunId(c.env, run.runId),
-    "https://agent-run.internal/browser-takeover/resume",
-    {
-      body: JSON.stringify(body),
-      headers: {
-        "Content-Type": "application/json",
-        "X-Cheatcode-User-Id": userId,
-      },
-      method: "POST",
-    },
+  return callAgentRun(
+    agentRunForRunId(c.env, run.runId).browserTakeoverResume(userId, body.takeoverId),
   );
 }