tensorlakeai · cooleel · Jun 15, 2026 · Jun 15, 2026
diff --git a/.claude-plugin/plugin.json b/.claude-plugin/plugin.json
@@ -1,6 +1,6 @@
 {
   "name": "tensorlake",
-  "version": "2.8.0",
+  "version": "2.9.0",
   "description": "Tensorlake SDK for agent sandboxes and sandbox-native orchestration. Use when building AI agents that need sandboxed execution environments, isolated tool calls, or durable workflow orchestration.",
   "author": "TensorLake",
   "homepage": "https://github.com/tensorlakeai/tensorlake-skills",

diff --git a/.github/scripts/sources.yaml b/.github/scripts/sources.yaml
@@ -25,28 +25,27 @@ sandbox_sdk.md:
     - https://docs.tensorlake.ai/sandboxes/commands.md
     - https://docs.tensorlake.ai/sandboxes/environment-variables.md
     - https://docs.tensorlake.ai/sandboxes/file-operations.md
-    - https://docs.tensorlake.ai/sandboxes/processes.md
     - https://docs.tensorlake.ai/sandboxes/networking.md
     - https://docs.tensorlake.ai/sandboxes/images.md
     - https://docs.tensorlake.ai/sandboxes/pty-sessions.md
     - https://docs.tensorlake.ai/sandboxes/docker.md
     - https://docs.tensorlake.ai/sandboxes/async.md
     - https://docs.tensorlake.ai/sandboxes/tunnels.md
-  sdk_version: "0.5.17"
-  last_verified: "2026-05-23"
+  sdk_version: "0.5.44"
+  last_verified: "2026-06-16"
 
 computer_use.md:
   sources:
     - https://docs.tensorlake.ai/sandboxes/computer-use.md
-  sdk_version: "0.5.17"
-  last_verified: "2026-05-23"
+  sdk_version: "0.5.44"
+  last_verified: "2026-06-16"
 
 sandbox_persistence.md:
   sources:
     - https://docs.tensorlake.ai/sandboxes/lifecycle.md
     - https://docs.tensorlake.ai/sandboxes/snapshots.md
-  sdk_version: "0.5.17"
-  last_verified: "2026-05-23"
+  sdk_version: "0.5.44"
+  last_verified: "2026-06-16"
 
 applications_sdk.md:
   sources:
@@ -73,8 +72,8 @@ applications_sdk.md:
     - https://docs.tensorlake.ai/applications/sandboxes.md
     - https://docs.tensorlake.ai/applications/guides/streaming-progress.md
     - https://docs.tensorlake.ai/applications/guides/logging.md
-  sdk_version: "0.5.0"
-  last_verified: "2026-04-24"
+  sdk_version: "0.5.44"
+  last_verified: "2026-06-16"
 
 integrations.md:
   sources:
@@ -84,8 +83,8 @@ integrations.md:
     - https://docs.tensorlake.ai/integrations/qdrant.md
     - https://docs.tensorlake.ai/integrations/databricks.md
     - https://docs.tensorlake.ai/integrations/motherduck.md
-  sdk_version: "0.5.0"
-  last_verified: "2026-04-24"
+  sdk_version: "0.5.44"
+  last_verified: "2026-06-16"
 
 platform.md:
   sources:
@@ -103,30 +102,32 @@ platform.md:
     - https://docs.tensorlake.ai/platform/playground/sample-documents.md
     - https://docs.tensorlake.ai/platform/security.md
     - https://docs.tensorlake.ai/platform/sso.md
-  sdk_version: "0.5.0"
-  last_verified: "2026-04-24"
+  sdk_version: "0.5.44"
+  last_verified: "2026-06-16"
 
 sandbox_usecases.md:
   sources:
     - https://docs.tensorlake.ai/sandboxes/skills-in-sandboxes.md
     - https://docs.tensorlake.ai/sandboxes/tool-calls.md
+    - https://docs.tensorlake.ai/sandboxes/claude-managed-agents.md
     - https://docs.tensorlake.ai/sandboxes/data-analysis.md
     - https://docs.tensorlake.ai/sandboxes/cicd-build.md
     - https://docs.tensorlake.ai/sandboxes/agentic-autoresearch.md
     - https://docs.tensorlake.ai/sandboxes/agentic-rl-reproducible-env.md
     - https://docs.tensorlake.ai/sandboxes/agentic-swarm-intelligence.md
+    - https://docs.tensorlake.ai/sandboxes/agentic-d&g.md
     - https://docs.tensorlake.ai/sandboxes/gspo-agentic-rl.md
     - https://docs.tensorlake.ai/sandboxes/chrome-cdp.md
     - https://docs.tensorlake.ai/sandboxes/harbor.md
     - https://docs.tensorlake.ai/sandboxes/remote-dev.md
-  sdk_version: "0.5.17"
-  last_verified: "2026-05-23"
+  sdk_version: "0.5.44"
+  last_verified: "2026-06-16"
 
 troubleshooting.md:
   sources:
     - https://docs.tensorlake.ai/applications/overview.md
     - https://docs.tensorlake.ai/applications/production/troubleshooting.md
     - https://docs.tensorlake.ai/document-ingestion/production/integration.md
     - https://docs.tensorlake.ai/document-ingestion/production/benchmarks.md
-  sdk_version: "0.5.0"
-  last_verified: "2026-04-24"
+  sdk_version: "0.5.44"
+  last_verified: "2026-06-16"
diff --git a/AGENTS.md b/AGENTS.md
@@ -1,5 +1,5 @@
 # Tensorlake SDK
-<!-- version: 2.8.0 -->
+<!-- version: 2.9.0 -->
 
 Tensorlake SDK — sandboxes for AI agents and applications. Use when the user mentions tensorlake or sandboxes, or asks about Tensorlake APIs/docs/capabilities. Also use when building an application, coding agent, or agentic system that needs a sandbox to run code — e.g., executing LLM-generated or untrusted code, persistence via suspend/resume, snapshots/checkpoints for forking parallel workers, custom images, exposing ports, egress allowlists, PTY/interactive shells, computer-use / desktop automation, browser automation (Chrome CDP, Playwright), local tunnels for non-HTTP protocols, async parallel sandboxes, Harbor evals or RL rollouts, file transfer, SSH access, remote-dev (VS Code Remote-SSH), or OCI base images. Also covers Tensorlake's sandbox-native durable workflow orchestration. Works alongside any LLM provider (OpenAI, Anthropic), agent framework (Claude/OpenAI agents SDK, LangChain), database, or API. When this skill applies, ALWAYS WebFetch https://docs.tensorlake.ai/llms.txt first.
 

diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -2,6 +2,21 @@
 
 All notable changes to the TensorLake skill are documented here.
 
+## [2.9.0] — SDK 0.5.44 — 2026-06-16
+
+### Added
+- **`references/sandbox_sdk.md`** — new "Managed Processes" section: supervised background processes via `start_process(..., name=, restart=RestartPolicyConfig(...), health_check=ProcessHealthCheck(...), user=)`, plus `get_process` / `restart_process` (TS `getProcess` / `restartProcess`), the new imports (`ProcessHealthCheck`, `ProcessHealthCheckType`, `RestartPolicy`, `RestartPolicyConfig`), restart policies (`never` / `on_failure` / `always`), HTTP/TCP health checks, default process user `tl-user`, and CLI `--detach` / `tl sbx ps|restart|kill`. New "Import an Image from a Registry" section (`tl sbx image import`, `import_sandbox_image()`, `importSandboxImage()`) and "Public Images" section (`--public` / `is_public` / `isPublic` for cross-namespace resolvable names).
+- **`references/sandbox_usecases.md`** — new "Claude Managed Agents" section (run the Claude Agent SDK / managed agents on sandboxes: brain-vs-hands model, three orchestrator modes, recreate-vs-resume for long sessions, per-command env injection, common failure modes) and a concise "Agentic Dungeons & Dragons" section (branch→map→reduce multi-agent demo running untrusted dice scripts in isolated sandboxes). Added `claude-managed-agents.md` and `agentic-d&g.md` to the source list.
+- **`references/feature_lookup.md`** — new entries for Claude managed agents and Agentic Dungeons & Dragons.
+
+### Changed
+- **`references/sandbox_sdk.md`** — bumped SDK version 0.5.17 → 0.5.44 and `Last verified:` to 2026-06-16. Removed `sandboxes/processes.md` from the source list (it now redirects into `sandboxes/commands.md`, "Commands & Processes"). Image build API now uses `image.build(registered_name=...)` / `registeredName` (was bare `build()`), with `context_dir` / `contextDir`, `builder_disk_mb` / `builderDiskMb`, corrected build defaults, and fully-qualified base names; added `tl sbx image register` / `tl sbx image ls`. `SandboxInfo` corrected: lowercase `SandboxStatus` enum, added `ingress_endpoint` / `sandbox_url` / `network`, nullable `image` / `timeout_secs` / `exposed_ports` / `entrypoint`, dropped undocumented `secret_names` and `resources.disk_mb`. "Supported Build Operations" rewritten to the live Dockerfile limitations. Expanded SSH (tmux/screen persistent shells, auth-failure messages, `tl sbx ls -r`, `IdentitiesOnly`). File ops use `bytes(read_file(...))` and `list_directory(...).entries`.
+- **`references/applications_sdk.md`** — bumped SDK version 0.5.0 → 0.5.44 and `Last verified:` to 2026-06-16. CLI corrected to `tl app deploy` / `tl app new`; `@application` defaults clarified (no retries by default, any-region); class-method applications callable by name string; `progress.update()` corrected to positional `current, total, message=None, attributes=None`; `File` raw-bytes / 5 TB note; scaling rate-limit formula (`max_containers × concurrency`); Cron Scheduler expanded (per-minute granularity, `schedule_id`, List/Delete endpoints, 1 MiB input limit); Secrets corrected to envelope encryption (per-project DEK wrapped by KMS KEK, mTLS); Observability/Logging corrected to structlog-style `Logger` usage, log levels TRACE–ERROR, 7-day retention, and the `GET /logs` query API.
+- **`references/sandbox_persistence.md`** — bumped SDK version 0.5.17 → 0.5.44 and `Last verified:` to 2026-06-16. Corrected the memory-snapshot restore lock: only image, resources (CPUs, memory), and entrypoint are inherited/immutable — `secrets` removed from that list (fixed in the TL;DR, Snapshot Types table, restore bullet, and Limitations). Clarified omitted snapshot type is a server-side default (currently `filesystem`).
+- **`references/computer_use.md`** — bumped SDK version 0.5.17 → 0.5.44 and `Last verified:` to 2026-06-16 (content already matched the live doc; no material changes).
+- **`.github/scripts/sources.yaml`** — bumped `sdk_version` / `last_verified` to `0.5.44` / `2026-06-16` for `sandbox_sdk.md`, `computer_use.md`, `sandbox_persistence.md`, `applications_sdk.md`, and `sandbox_usecases.md`. Removed `processes.md` from `sandbox_sdk.md` sources; added `claude-managed-agents.md` and `agentic-d&g.md` to `sandbox_usecases.md` sources.
+- **`README.md`** — illustrative source-header example bumped to `tensorlake 0.5.44` / `2026-06-16`.
+
 ## [2.8.0] — SDK 0.5.17 — 2026-05-23
 
 ### Added

diff --git a/README.md b/README.md
@@ -180,8 +180,8 @@ Each reference file has a source header that tracks which doc pages it was built
 Source:
   - https://docs.tensorlake.ai/sandboxes/lifecycle.md
   - https://docs.tensorlake.ai/sandboxes/commands.md
-SDK version: tensorlake 0.5.17
-Last verified: 2026-05-23
+SDK version: tensorlake 0.5.44
+Last verified: 2026-06-16
 -->
 ```
 

diff --git a/SKILL.md b/SKILL.md
@@ -19,7 +19,7 @@ description: >
   https://docs.tensorlake.ai/llms.txt first.
 metadata:
   author: tensorlake
-  version: 2.8.0
+  version: 2.9.0
 ---
 
 # What can you do with Tensorlake SDK

diff --git a/references/applications_sdk.md b/references/applications_sdk.md
@@ -23,8 +23,8 @@ Source:
   - https://docs.tensorlake.ai/applications/sandboxes.md
   - https://docs.tensorlake.ai/applications/guides/streaming-progress.md
   - https://docs.tensorlake.ai/applications/guides/logging.md
-SDK version: tensorlake 0.5.0
-Last verified: 2026-04-24
+SDK version: tensorlake 0.5.44
+Last verified: 2026-06-16
 -->
 
 # TensorLake Applications SDK Reference
@@ -59,7 +59,7 @@ from tensorlake.applications import (
     get_remote_request,
     Future, RETURN_WHEN,
     RequestContext, Request, File, Image, Retries,
-    ReplayMode, RequestError,
+    ReplayMode, RequestError, Logger,
 )
 ```
 
@@ -72,8 +72,8 @@ Entry point decorator. Must wrap a function also decorated with `@function()`.
 ```python
 @application(
     tags: dict[str, str] = {},
-    retries: Retries | None = None,
-    region: str | None = None,               # "us-east-1" or "eu-west-1"
+    retries: Retries | None = None,          # Default for all functions; no retries by default
+    region: str | None = None,               # "us-east-1" or "eu-west-1"; default any region
 )
 ```
 
@@ -116,6 +116,9 @@ class MyProcessor:
     @function(gpu="T4")
     def process(self, data: str) -> str:
         return self.model.predict(data)
+
+# Call a class-method application by name:
+# run_remote_application("MyProcessor.process", data="...")
 ```
 
 ## Calling Functions
@@ -228,8 +231,11 @@ output = request.output()
 ```
 
 ```bash
+# Scaffold a new application (creates hello_world/hello_world.py)
+tl app new hello_world
+
 # Deploy before running remotely
-tl deploy path/to/app.py
+tl app deploy path/to/app.py
 ```
 
 ## Durable Execution
@@ -266,13 +272,20 @@ ctx.state.set(key, value)
 ctx.state.get(key, default=None)
 
 # Metrics
-ctx.metrics.timer(name, value)
-ctx.metrics.counter(name, value)
+ctx.metrics.timer(name, value)         # duration in seconds
+ctx.metrics.counter(name, value=1)     # counters start at 0
 
 # Progress reporting (also resets timeout)
-ctx.progress.update(current=10, total=100, message="Processing...")
+ctx.progress.update(
+    current,                           # int | float (step or percentage)
+    total,                             # int | float (total steps, or 100 for percent)
+    message=None,                      # optional str
+    attributes=None,                   # optional dict[str, str] metadata
+)
 ```
 
+Poll progress at `GET /applications/{app}/requests/{request_id}/progress`.
+
 ## Image Builder
 
 Build custom container images for functions.
@@ -303,6 +316,10 @@ file.content       # bytes
 file.content_type  # str
 ```
 
+A `File` arg/return bypasses JSON/pickle and is passed as raw bytes. As an application
+input/output, the HTTP body is the raw bytes and content type. Files up to 5 TB are
+supported (loaded fully into memory; no lazy loading yet).
+
 ## Retries
 
 ```python
@@ -328,29 +345,44 @@ def agent(prompt: str) -> str: ...
 ```
 
 - `warm_containers`: Pre-warmed containers for zero cold-start latency
-- `max_containers`: Upper limit; excess requests queued FIFO
+- `max_containers`: Upper limit; excess requests queued FIFO (no separate queue infra needed)
 - `concurrency`: Concurrent requests per container
 - Default (no params): scales dynamically from zero, no upper limit
+- **Rate-limit external APIs**: total concurrent calls = `max_containers` × `concurrency`
 
 ## Cron Scheduler
 
-Schedule periodic application runs via REST API:
+Schedule recurring invocations of a deployed application via REST API (or the UI).
+Schedule starts immediately on creation.
 
 ```python
 import requests, base64, json
 
-payload = {"cron_expression": "0 * * * *"}
+payload = {"cron_expression": "0 * * * *"}   # 5-field cron; "* * * * *" = every minute
 input_data = json.dumps({"report_type": "daily"}).encode()
-payload["input_base64"] = base64.b64encode(input_data).decode()
+payload["input_base64"] = base64.b64encode(input_data).decode()  # optional, max 1 MiB decoded
 
-response = requests.post(
+# Create
+resp = requests.post(
     f"https://api.tensorlake.ai/applications/{application}/cron-schedules",
     json=payload,
     headers={"Authorization": "Bearer TENSORLAKE_API_KEY"},
 )
+schedule_id = resp.json()["schedule_id"]   # save it; needed to delete
+
+# List  -> {"schedules": [{id, application_name, cron_expression, next_fire_time_ms,
+#                          last_fired_at_ms, created_at, enabled}, ...]}
+requests.get(f"https://api.tensorlake.ai/applications/{application}/cron-schedules", headers=...)
+
+# Delete (permanent; to modify, delete + recreate)
+requests.delete(
+    f"https://api.tensorlake.ai/applications/{application}/cron-schedules/{schedule_id}",
+    headers=...,
+)
 ```
 
-Minimum interval: 60 seconds. Max 100 schedules per application.
+Minimum interval: 60 seconds (sub-minute expressions rejected with 400). Max 100 schedules
+per application. Max input payload 1 MiB decoded.
 
 ## Exceptions
 
@@ -374,15 +406,23 @@ def my_func() -> str:
     key = os.environ["OPENAI_API_KEY"]
 ```
 
-Redeploy applications after updating secrets. AES-256-GCM encryption, in-memory decryption only during execution.
+Redeploy applications after adding/updating a secret for new values to take effect.
+Envelope encryption (AES-256-GCM): a per-project DEK wrapped by a KEK in AWS KMS;
+decrypted in-memory only on dataplane machines running the function, over mTLS.
 
 ## Observability
 
-Every `@function()` call is automatically traced. The dashboard shows function call sequence, timing (including cold starts), dependency visualization, and status. Use standard Python `logging` module; logs are captured automatically.
+Every `@function()` call is automatically traced. The dashboard shows function call sequence, timing (including cold starts), dependency visualization, and status. Use standard Python `print()`/`logging` (captured automatically, default level INFO), or the built-in structured logger:
 
 ```python
 from tensorlake.applications import Logger
 
 logger = Logger.get_logger(module="my_app")
-logger.info("starting run", log_attributes={"request_id": "req-123"})
+logger.info("User logged in", user_id=123)     # structlog-style kwargs
+logger.error("An error occurred", exc_info=True)
+logger = logger.bind(request_id="req-123")      # bind context to all subsequent logs
 ```
+
+Levels TRACE(1)..ERROR(5). `structlog` (JSON renderer) also supported. Logs retained 7 days
+(extendable to 30 days / 1 year). Query via `GET /applications/{app}/logs` with filters
+(`requestId`, `function`, `level`, `gate=and|or`, `head`/`tail`, `nextToken`).
diff --git a/references/computer_use.md b/references/computer_use.md
@@ -1,8 +1,8 @@
 <!--
 Source:
   - https://docs.tensorlake.ai/sandboxes/computer-use.md
-SDK version: tensorlake 0.5.17
-Last verified: 2026-05-23
+SDK version: tensorlake 0.5.44
+Last verified: 2026-06-16
 -->
 
 # Computer Use (Desktop Automation)

diff --git a/references/feature_lookup.md b/references/feature_lookup.md
@@ -21,6 +21,7 @@ Bundled offline equivalent of [docs.tensorlake.ai/llms.txt](https://docs.tensorl
 - [Computer use / desktop automation](computer_use.md) — XFCE + Firefox, screenshots, mouse/keyboard, noVNC live view
 - [Drive Chrome over CDP](sandbox_usecases.md#drive-chrome-over-cdp) — sandboxed Google Chrome with `--remote-debugging-port`, Playwright `connect_over_cdp`, raw CDP WebSocket, `chrome-devtools-mcp` for Claude Code / Codex
 - [Skills in sandboxes](sandbox_usecases.md) — bundling Claude Code, Codex, Cursor, Cline, Windsurf, GitHub Copilot, Google ADK skills inside images
+- [Claude managed agents](sandbox_usecases.md#claude-managed-agents) — run Claude Agent SDK / managed agents on sandboxes; orchestrator modes, recreate-vs-resume for long sessions, per-command env injection
 - [SSH access](sandbox_sdk.md#ssh) — connect with `ssh`/`scp`/`sftp`/`rsync`, port forwarding (`-L` / `-D` / `-R`), VS Code Remote-SSH and JetBrains Gateway
 - [Sandbox as a dev environment](sandbox_usecases.md#sandbox-as-a-dev-environment) — portable cloud workstation with idle-suspend, resume-by-name, and persistent `~/.vscode-server`
 - [OCI base images](sandbox_sdk.md#base-images) — build from any standard OCI reference (`python:3.12-slim`, `node:22-alpine`, `ghcr.io/...`) plus `tensorlake/*` bases; private-registry auth via `~/.docker/config.json`
@@ -30,6 +31,7 @@ Bundled offline equivalent of [docs.tensorlake.ai/llms.txt](https://docs.tensorl
 
 - [Tool calls / sandbox-as-tool](sandbox_usecases.md#ai-code-execution) — LLM code-execution tool, executing untrusted/LLM-generated code with network policy
 - [Agentic swarm intelligence](sandbox_usecases.md) — fan-out parallel specialist agents
+- [Agentic Dungeons & Dragons](sandbox_usecases.md#agentic-dungeons--dragons) — branch→map→reduce multi-agent demo running untrusted dice scripts in isolated sandboxes
 - [RL training (GSPO)](sandbox_usecases.md) — fine-tune on code generation with sandbox reward oracle
 - [RL reproducible environments](sandbox_usecases.md) — deterministic isolated rollouts
 - [Agentic autoresearch loop](sandbox_usecases.md) — overnight ML script self-improvement with parallel sandbox races