Skip to content

agent-creation-permissions-copy.md

Latest commit

 

History

History
147 lines (116 loc) · 4.39 KB

agent-creation-permissions-copy.md

File metadata and controls

147 lines (116 loc) · 4.39 KB

Agent creation permissioning copy (pre-removal)

Source: 2e33f60 removed guided creation files and the copy came from its parent tree (2e33f60^).

src/features/agents/components/AgentCreateModal.tsx copy

Step headers

  • Define Ownership
  • Assign full accountability.
  • Set Authority Level
  • Define how independently this agent can act.
  • Launch Agent
  • Review mandate and activate.
  • Activation begins immediately.
  • Name reflects its role. You can change it later.
  • You can adjust ownership and authority later.
  • Authority can be configured after launch.

Authority profiles

  • Conservative

  • Acts with review required.

  • Can modify code and files directly

  • Code and files access is on by default

  • No automatic system actions

  • Collaborative

  • Acts with approval.

  • Can modify code and files directly

  • Runs system actions with approval

  • Uses web access for context

  • Autonomous

  • Acts independently.

  • Can modify your codebase directly

  • Can operate your system automatically

  • Uses web access while iterating

Authority summary text

  • Autonomous - acts independently
  • Collaborative - acts with approval
  • Conservative - acts with review required

Authority configuration UI copy

  • Web Access
  • Internet research and fetch tools.
  • Code and Files
  • Controls codebase and file modification behavior.
  • Can modify your codebase directly.
  • System Access
  • Controls system actions and command execution.
  • Can operate your system automatically.
  • Off
  • On
  • Ask first
  • Auto
  • Can run commands automatically without approval prompts.

Step/action and control labels

  • Fine-tune capabilities (optional)
  • Hide fine-tune capabilities
  • Additional capabilities (optional) (button label logic in this version)
  • Recommended
  • Conservative autonomy profile
  • Collaborative autonomy profile
  • Autonomous autonomy profile
  • Web access off
  • Web access on
  • File changes on
  • Command execution off
  • Command execution ask first
  • Command execution auto

Modal summary section

  • This agent will be accountable for:
  • This agent will:
  • Authority:
  • On launch it will:
  • What does this agent fully own?
  • This agent will be accountable for:

Other launch text

  • Agent name
  • You can rename this agent later in settings.
  • You can rename or change the avatar later.
  • You can rename this agent later.
  • Choose avatar
  • Shuffle
  • Launch agent
  • Launching...
  • You can adjust ownership and authority later.
  • This agent will own: (from summary card format, e.g. title + description)

src/features/agents/creation/compiler.ts copy

Starter and template copy (permission-relevant intent)

  • Evidence-first synthesis with broad access defaults.
  • Safe code changes with broad execution defaults.
  • High-autonomy coding with broad execution permissions.
  • Campaign drafting defaults with broad access.
  • Follow-up and planning support with broad defaults.
  • General-purpose baseline with broad defaults.

Boundaries / control language

  • Do not publish, send, or launch externally without approval.
  • Evidence beats intuition when stakes are non-trivial.
  • Unknowns should be visible, not hidden.
  • A concise synthesis is more useful than a long dump.
  • Correctness and reversibility come before speed.
  • Small scoped changes reduce operational risk.

Control mode defaults

  • Agent name is required.
  • Auto exec cannot be enabled when approval security is set to deny.
  • Auto exec requires runtime tools to be enabled.

Capability summary copy

  • Persona files: custom IDENTITY.md + SOUL.md; AGENTS.md remains the gateway default.
  • Web access is enabled for search and fetch tools.
  • Web access is disabled.
  • Can apply file edits directly on the host filesystem.
  • Sessions run without sandbox isolation.
  • Command execution is disabled.
  • Can run commands automatically without approval prompts.
  • Can run commands with approval prompts.

Execution role/profile summaries generated by compile

  • No command execution. (from conservative mapping in inspect/derived role view in same era)
  • Commands require approval.
  • Can operate your system automatically. (mode summaries and tooltips)

Notes

  • This is a copy extraction for restoration/reuse, not a diff.
  • The live strings came from the commit that removed the guided creation flow.