Software Engineer

Autonomous coding agent for repositories — explore codebases, implement changes, and run engineering tasks.

Try it free

development

Check Sandbox Status
Check sandbox status and verify the development environment is ready. Always call this first to ensure the sandbox is initialized.
Claude Code
Delegate a coding task to Claude Code, an expert AI coding agent running on the sandbox. Claude Code has full access to the filesystem, can read/write/edit files, run shell commands, search, fetch web content, and work autonomously. All permissions are auto-approved (the sandbox is the security boundary). Use this as your primary tool for all code changes, file creation, refactoring, debugging, and implementation work. The prompt should be specific and self-contained -- Claude Code does not have context from previous calls unless you set continue_session=true to resume the most recent session in the working directory. Output streams in real-time to the web terminal so the user can watch progress. The final text result is returned when Claude Code finishes. Model selection is automatic: in fast mode every call runs on Sonnet with Opus as fallback; in max mode pass `task_type="plan"` for heavy architectural work (Opus, Sonnet fallback) and leave `task_type` as the default `implement` for code changes (Sonnet, Opus fallback). Use system_prompt_append to inject project-specific context (tech stack, conventions, coding patterns) so Claude Code follows the right patterns on the first attempt.
Github Cli
Execute GitHub CLI (gh) commands. Supports all GitHub operations through a single unified interface. Common operations: Repositories: - List repos under a known org: "repo list <org-name> --limit 100 --json name,owner,url" ← prefer this - List repos for the authenticated user: "repo list <login> --limit 100 --json name,owner,url" ← prefer this - Scoped search: "search repos <reponame> org:<org-name> --json fullName,url --limit 10" - Scoped search (user): "search repos <reponame> user:<login> --json fullName,url --limit 10" - Unscoped public search (LAST RESORT): "search repos <reponame> --json fullName,url --limit 10" - View repo: "repo view owner/repo --json description,stars,forks" - Clone info: "repo view owner/repo --json sshUrl,defaultBranch" When a user provides only a repo name (not owner/repo), do NOT guess the owner and do NOT lead with unscoped "search repos <name>" — that hits GitHub's global public index and returns noisy, unrelated repos. Instead, read the orgs and authenticated login from the <github_context> block in your system prompt and run "repo list <org>" / "repo list <login>" (or scoped "search repos <name> org:<org>" / "user:<login>") FIRST. Only fall back to unscoped search if every scoped query returned zero matches. If still ambiguous, ask the user to confirm the full owner/repo or paste the URL — never clone an unrelated public repo just because it shares a name. Pull Requests: - List PRs: "pr list --repo owner/repo --state open --json number,title,author" - View PR: "pr view 123 --repo owner/repo --json title,body,state,reviewDecision" - Get diff: "pr diff 123 --repo owner/repo" - Create PR: "pr create --title 'feat(scope): short summary' --body $'## Summary\nWhat this PR does and why.\n\n## Changes\n- Key change\n\n## Testing\n- How it was tested' --base main" - Review PR: "pr review 123 --approve --body 'LGTM'" - Request changes: "pr review 123 --request-changes --body 'Please fix...'" - Comment: "pr comment 123 --body 'Question about...'" - Checkout PR: "pr checkout 123" - Merge PR: "pr merge 123 --squash --delete-branch" - List files: "pr view 123 --json files" Issues: - List issues: "issue list --repo owner/repo --state open --json number,title,labels" - View issue: "issue view 42 --repo owner/repo --json title,body,comments" - Create issue: "issue create --title 'Bug' --body 'Description' --label bug" - Close issue: "issue close 42 --repo owner/repo" Other: - Releases: "release list --repo owner/repo" - Workflows: "workflow list --repo owner/repo" - API calls: "api repos/owner/repo/pulls/123/comments" Always use --json flag when you need structured data.
Glob File Search
Find files matching a glob pattern, sorted by modification time. Use this for pattern-based file discovery when you need to find files by name patterns rather than content. Examples: - Find all TypeScript test files: "*.test.ts" - Find all components: "**/components/**/*.tsx" - Find all config files: "**/*.config.{js,ts}"

Example use cases

What teams automate with Software Engineer

7 realistic workflows that combine Software Engineer with your other apps.

Engineering
When Linear assigns a backend task, use Software Engineer to scaffold the API in GitHub, test via Heroku preview, and open a PR.
Engineering
For each failing GitHub check, invoke Software Engineer to patch, run Bitrise locally, and update the Linear ticket.
Ops
When Cloudflare Workers errors spike, use Software Engineer to patch bindings, deploy, and confirm in Sentry.
Security
When Sentry reports an auth bypass, use Software Engineer to add tests, fix in GitHub, and notify security in Slack.
Product
When Shortcut specs lack API detail, use Software Engineer to draft OpenAPI in Notion, implement in GitHub, and demo in the web browser.
Data
When Snowflake pipelines stall, use Software Engineer to fix the worker in AWS, replay jobs, and validate in Metabase.
Engineering
After Firebase rule changes, use Software Engineer to update clients, run Bitrise, and ship to Google Play internal testing.

Similar apps

Explore other integrations teams often connect with Software Engineer.

GitHubdevelopment

Herokudevelopment

Linearproject management

App Store Connectdevelopment

Back to App Directory