Conversation
…ats (#33) * feat: accept text format for sandbox nodeDependencies input Switch nodeDependencies from a JSON object editor to a textarea, matching the look-and-feel of pythonRequirementsTxt and envVars. Inputs are parsed as either npm CLI-style lines (one `package@version` per line, scoped packages and `latest`-default supported) or a JSON object for back-compat. Adds unit tests for parseNodeDependencies under tests/unit/ and updates AGENTS.md with guidance for adding future unit tests. https://claude.ai/code/session_011t31rBsLpeNZBwUUsPtFkV * chore: add commented example to nodeDependencies prefill Mirrors the style of initShellScript so the textarea hints at the expected line format on first open. https://claude.ai/code/session_011t31rBsLpeNZBwUUsPtFkV * docs: explain parseJsonObject coercion and graceful failure https://claude.ai/code/session_011t31rBsLpeNZBwUUsPtFkV --------- Co-authored-by: Claude <noreply@anthropic.com>
Adds a new `mcpConnections` array input (resourceType: mcpConnector,
wildcard `*` server URL) that lets the user pick any MCP Connector they
have authorized in Settings > API & Integrations. On sandbox start the
Actor writes the chosen Connector IDs to /sandbox/mcp.json in the
standard `mcpServers` shape so tools like `mcpc connect` can pick them
up immediately. Authorization uses a `${APIFY_TOKEN}` placeholder so
the secret never lands on disk.
Co-authored-by: Claude <noreply@anthropic.com>
* Add actor page link to shell welcome banner and fix Venv alignment Surface the running actor's top-level URL in the shell welcome screen so users can easily navigate from /shell/ back to the landing page. Also realign the Venv row so it matches the column used by the other System Info entries. * Shrink ASCII banner and add run-details link to welcome screen Halve the Apify logo (27 → 14 lines, ~50 → 27 chars wide) so the welcome screen takes up less vertical space when opening the shell. Also add a Documentation entry pointing at the Apify console run view, derived from ACTOR_RUN_ID so users can jump to the run details page from inside the terminal. --------- Co-authored-by: Claude <noreply@anthropic.com>
* Align /llms.txt and / content Both endpoints now cover the same sections in the same order: Quick links, MCP, Code execution, Filesystem, Proxy Mappings, Response format, Working directories, Configuration. - llms.txt: add Proxy Mappings section; replace SDK code examples with curl commands matching the landing page - landing page: add collapsible Response format, Working directories, and Configuration cards https://claude.ai/code/session_01Ey8UdUPzR4ZbcLssnoZKis * Generate llms.txt from the landing page HTML Replace the standalone llms.md template with HTML-to-Markdown conversion of the rendered landing page. The landing page is now the single source of truth for the content of both endpoints. - Use node-html-markdown + node-html-parser to convert landing HTML to MD - Custom <pre> translator reads data-lang for fenced code language - Strip script/style and elements marked data-no-md, plus copy/collapse buttons and status badge before conversion - Hero h1 becomes "Apify AI Sandbox" (the eyebrow + "Landing page" h1 are collapsed into a single proper title) - Quick Links card now contains a visible endpoint list rendered in both HTML and MD; action buttons are kept for HTML and marked data-no-md - Card titles unified at h2 and example labels at h3 (with CSS to keep the previous visual sizing) so the heading hierarchy is consistent in MD - Pre blocks have a data-lang attribute so fences render with proper language tags https://claude.ai/code/session_01Ey8UdUPzR4ZbcLssnoZKis --------- Co-authored-by: Claude <noreply@anthropic.com>
* Add commented example prefills for setup input fields Replace the active `apt-get update` line in initShellScript prefills (and empty pythonRequirementsTxt prefill) with fully commented-out examples so the defaults have no effect on the run. Users can now uncomment lines or add their own. https://claude.ai/code/session_0115MxnEh2NxRT2cfkNjGEfu * Use stable @apify/mcpc instead of @beta Named connection support has graduated to the stable release, so the beta pin is no longer needed. https://claude.ai/code/session_0115MxnEh2NxRT2cfkNjGEfu * Add CLAUDE.md with PR description convention Keep PR descriptions to 2-4 bullets; the diff is the record of what changed, the description explains intent. https://claude.ai/code/session_0115MxnEh2NxRT2cfkNjGEfu --------- Co-authored-by: Claude <noreply@anthropic.com>
The browser terminal (ttyd) only displayed "Press ⏎ to Reconnect" when the container stopped, leaving users without any explanation. Broadcast a banner to open terminal sessions (via the PTY slave devices ttyd reads from) before exiting, covering idle-timeout, migration, and abort. The idle-timeout path adds a brief flush delay so the banner reaches the browser before the socket drops. Co-authored-by: Claude <noreply@anthropic.com>
Apify's run Live View always loads the container root, so serve a full-page iframe of /shell/ at `/` to surface the terminal directly in the run console. Move the docs/API landing page to /info and update the in-app links, welcome script, and output schema accordingly. https://claude.ai/code/session_01YJZHMbJgwvYNc5WRf8TJ6v Co-authored-by: Claude <noreply@anthropic.com>
* Make input schema prefills and help texts more concise Trim the verbose, multi-line prefill templates added in #39 down to a short hint plus one or two commented examples, and collapse the longer field descriptions (node/python deps, init script, env vars, idle timeout, MCP connectors) into single sentences. The sandbox keeps its distinct env-vars wording since it deletes them after the init script. https://claude.ai/code/session_01WZCxqZajjnGHGFiqy4xUec * Bump default idle timeout to 15 minutes Raise the idleTimeoutSeconds default from 600s (10 min) to 900s (15 min) for the sandbox, claude-code, and opencode actors, updating the schema default/prefill, the sandbox code fallback, JSDoc, and READMEs. OpenClaw stays at 0 (idle shutdown disabled), which already exceeds 15 minutes. https://claude.ai/code/session_01WZCxqZajjnGHGFiqy4xUec * Change skills input to textarea (line + JSON-array formats) Switch the skills input from a stringList to a textarea across all four actors. A new parseSkills() accepts one skill per line (e.g. anthropics/skills) with # comments and blank lines ignored, a JSON array, or a legacy string array (saved stringList task inputs), de-duplicating the result. Adds unit tests and updates the schemas, types, and docs. https://claude.ai/code/session_01WZCxqZajjnGHGFiqy4xUec * Extract shared safeParseJson helper Pull the try/JSON.parse/log-warning/shape-check scaffolding from env-vars.ts, node-deps.ts, and skills.ts into safe-json.ts as safeParseJson + isFlatJsonObject. Each caller now passes a predicate and field-scoped shape description; behavior is unchanged. Adds 20 unit tests for the new helper. https://claude.ai/code/session_01WZCxqZajjnGHGFiqy4xUec --------- Co-authored-by: Claude <noreply@anthropic.com>
* Use box emoji as favicon Replace the binary favicon.ico with an inline SVG data URI rendering the 📦 emoji, and drop the now-unused /favicon.ico endpoint. * Apply PR description convention to externally-opened PRs Make the convention bind even when a PR is created from the Claude Code UI, so the description gets trimmed instead of left as the default Summary / Key Changes / Implementation Details template. --------- Co-authored-by: Claude <noreply@anthropic.com>
Client-side SPA at /browse and /browse/<path> that fetches the existing /fs/* JSON endpoints to render directory listings and inline file previews (text / image, with raw + ZIP download links). Adds a Browse files quick link to /info. Co-authored-by: Claude <noreply@anthropic.com>
* Rename inputs and product name, clarify nodeDependencies format - "Apify AI Sandbox" → "Apify AI Code Sandbox" across UI, READMEs, logs - Drop the eyebrow + "Landing page" stub; promote the product name to H1 - Input: pythonRequirementsTxt → pythonRequirements - Input: idleTimeoutSeconds → idleTimeoutSecs (sandbox + sub-actors) - nodeDependencies description: note we accept the same syntax as `npm install <pkg>` https://claude.ai/code/session_01Vd2LZY2wcnUKayh5anMsAe * Add /shell?launch=<cmd>, rename mcpConnectors, polish landing page - /shell?launch=<cmd> translates to ttyd's arg=-c, arg=source bashrc; <cmd> for HTTP + WebSocket; works for arbitrary commands, not just agents - Sub-actor output URLs now use ?launch=claude / ?launch=opencode / ?launch=openclaw%20tui (instead of the long arg= form) - Quick links: replace the agent buttons with ?launch=, add an "Actor on Apify" link to https://apify.com/apify/ai-sandbox - Input: mcpConnections → mcpConnectors (matches the platform's "MCP Connectors" terminology) - Landing/llms: "Code execution" → "Code execution API", "Filesystem endpoints" → "Filesystem API", "Proxy Mappings" → "Proxy mappings", "Add Mapping" → "Add mapping" - Proxy mappings: explain what they're for (in landing copy and the actor.json input description) https://claude.ai/code/session_01Vd2LZY2wcnUKayh5anMsAe * Install agent CLIs via npm and add Codex Switch Claude Code and OpenCode from curl install scripts to npm-global, which gives the same binaries (Claude uses Node as launcher; OpenCode bundles Bun). Also add @openai/codex — a self-contained Rust binary delivered the same way. Single install mechanism, version-pinnable, no more curl pipelines. Welcome banner and the build-time version capture now include Codex; the landing page Quick links gain a Codex button. https://claude.ai/code/session_01Vd2LZY2wcnUKayh5anMsAe * Add codex sub-actor Mirrors the claude-code / opencode metamorph pattern: a thin wrapper that metamorphs into the main AI Code Sandbox and lands the user in a terminal running codex (output URL: /shell?launch=codex). Codex CLI itself is already installed in the main sandbox image. https://claude.ai/code/session_01Vd2LZY2wcnUKayh5anMsAe * Revert sub-actor changes; defer to a follow-up PR Drops the claude-code/ and opencode/ updates and the new codex/ sub-actor from this PR so they can be redone as a focused follow-up. The /shell?launch= plumbing in the main sandbox stays, so the new URLs will Just Work once the sub-actors are updated separately. https://claude.ai/code/session_01Vd2LZY2wcnUKayh5anMsAe * Revert openclaw changes; defer to a follow-up PR --------- Co-authored-by: Claude <noreply@anthropic.com>
Reverts #43: serve the docs/API landing page at `/` again instead of a full-page iframe of /shell/. Removes the `/info` route, the `docsUrl` output property, and the now-unused live-view template, and repoints the internal links back to `/`. https://claude.ai/code/session_01Xxnc9G8bTasCXR7HhSVrJG Co-authored-by: Claude <noreply@anthropic.com>
* Change default terminal username from apify to actor Update the PS1 prompt in the sandbox shell template so the terminal shows actor@sandbox instead of apify@sandbox. * Auto-approve AI agents and fix /shell launch restart loop - /shell?launch=<cmd> now runs in a persistent interactive shell that echoes the command and surfaces a non-zero exit, instead of a bash -c that exits and makes ttyd respawn it in a restart loop. - Claude: wrap the command with --dangerously-skip-permissions (settings bypass mode shows a blocking confirmation dialog); pre-accept onboarding. - Codex: ~/.codex/config.toml with approval_policy=never + danger-full-access. - OpenCode: permission "allow" so nothing prompts. - OpenClaw: open both YOLO policy layers (tools.exec + exec-approvals.json). - Point the wrapper actors' launch URLs at ?launch= so they get the fix. --------- Co-authored-by: Claude <noreply@anthropic.com>
* feat(landing): redesign landing page, remove Actor on Apify button Refine to a calmer dark theme with a single accent and unified quick-link buttons, and fix inline code chips that rendered as misaligned full-width blocks. https://claude.ai/code/session_01JbUM5pBzpZRodXuWN4C7vf * refactor(landing): extract page styles into landing.css Move the inline <style> block into a dedicated landing.css and inject it at render time via EJS, keeping the page self-contained (no extra request/route). The build copies *.css to dist/templates alongside the .ejs. https://claude.ai/code/session_01JbUM5pBzpZRodXuWN4C7vf --------- Co-authored-by: Claude <noreply@anthropic.com>
* Deliver shutdown banner directly over the terminal WebSocket The terminal is proxied (browser <-> Actor <-> ttyd), so writing the banner to the PTY and relaying it back through ttyd lost the race against process exit on platform-initiated stops (abort, migration, run timeout) — only the idle path's flush delay reliably worked. Write a ready-made ttyd output frame straight to the browser socket instead, so the bytes reach the kernel before the proxy dies; fall back to the PTY devices when no socket is tracked. https://claude.ai/code/session_01KrCLcsSMPpsyWwuTyCzYv8 * Pin ttyd to 1.7.7 The shutdown banner writes ttyd's own WebSocket output frames directly to the browser, so an unpinned ttyd could change its wire protocol and silently break banner rendering. Pin it and document how to re-check on a future bump. https://claude.ai/code/session_01KrCLcsSMPpsyWwuTyCzYv8 --------- Co-authored-by: Claude <noreply@anthropic.com>
The skills CLI accepts a repo URL as well as the owner/repo shorthand, and parseSkills already passes both through unchanged. Surface this in the input description, prefill example, README, and doc comments, and add tests covering URL pass-through (line + JSON formats). https://claude.ai/code/session_016XgcmN1VU9uvoaAcD1rdLF Co-authored-by: Claude <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.