Customize your Self folder before download.

Model creator The company or lab that made the model, such as Anthropic, OpenAI, or Google. Model provider The service currently providing access to the model. This may match the creator, but it is not the app or harness. Model name Example: Claude Sonnet 4.5, GPT-5.2, Gemini 3 Pro, or unknown. Model version or release date Use a version, date, or release label if known. Model release notes URL Model card URL Benchmark or eval URL Known capabilities List only capabilities disclosed by the provider, service, or user setup. Known limits List model limitations, uncertainty, or things the assistant should not claim.

Harness name The app, site, CLI, IDE, desktop client, third-party wrapper, or custom runtime the user is actually using. Harness provider or operator Example: Anthropic, OpenAI, Google, Cursor, a company team, or local custom setup. Interface type Unknown Website Desktop app CLI IDE extension Third-party wrapper Custom harness This is the access interface, not the model name. Access details Optional URL, app name, command name, workspace behavior, account scope, or launch notes. Available surfaces List the app surfaces the assistant may see, such as files, browser, terminal, or desktop controls. Capabilities and integrations List safe harness features, APIs, endpoints, connectors, or state snapshots the assistant may inspect or invoke. Script and tool routing Map common requests to tools or files, such as browser retry to scripts/browser.md. Permission boundaries State what requires confirmation, what cannot be assumed, and how to fail safely. Known harness limits List limits from the app, permissions, shell, browser, APIs, sandbox, or available tools.

Default routing workflow Declare how the system chooses the next file, tool, model, subagent, or fallback when a request arrives. Context loading workflow Keep this compact. The assistant should start with SELF.md and open only relevant slots. Tool mediation workflow Explain how tool use is coordinated with harness permissions and script routes. Model or subagent selection workflow Optional. Use this when the harness can choose models, subagents, modes, or specialists. Private context workflow State when private files such as user.md or chrono/ may participate. Harness route summary Repeat the main harness routes here so orchestration remains useful even if users skip this file. Model and harness adaptation notes Optional notes about disclosed model or harness strengths, limits, quirks, and habits to navigate. Fallback workflow Explain what to do when a slot, route, tool, script, model, or subagent is missing. User-facing workflow defaults Broad process defaults only. Personal style and preferences still belong in user.md.

Personal biography Example: name, background, location, family details, or anything personal the assistant should not infer. Professional work context Example: field, job, professional acumen, current projects, recurring constraints, or places where the user wants help filling gaps. Communication preferences Example: Verbosity, tone, specificity, formatting, and how direct the assistant should be. Include a Beliefs And Worldview section Beliefs and worldview Sensitive context. Add only user-stated beliefs, values, philosophical preferences, or answer boundaries.

Package name A short public label for this self/ package. It is not a Skill trigger. Customized from Self version The Self standard version this package was generated from. Created date Installed date Leave blank if the package has not been installed yet. Generator version Optional maintainer label Use a name, handle, or leave it blank. Do not put private facts here.

Manual edits in this preview are included in the ZIP for that file and do not change the wizard fields above. Regenerate rebuilds the selected file from the wizard answers above and discards review-editor edits for that file.