An agent profile is the saved who of a run: the agent and model it uses, how
hard it reasons, how many turns it may take, which MCP tiers and memories it may
reach, and which checks it must pass. This page lists every field a profile
carries. For the walk-through — creating one, promoting the workspace default,
curating MCPs and skills — read
Create and manage an agent profile.
Profiles live under Settings → Agents → Agent Profiles. The tab’s detail form
carries the everyday fields; the advanced editor (/settings/profiles/<id>,
reached from the row menu or the Advanced editor button) carries all of them,
with a Raw view for the JSON-only knobs and a Versions view of the runbook
history. Creating, editing, deleting and promoting a profile are
workspace-admin actions; a member sees the tab read-only.
Claude Code, Codex, Gemini, Qwen Code or OpenRouter — a provider-backed agent only.
Claude Code
Model policy
Auto (SupaCloud may choose the model from selection signals when the task sets none), Manual (the task decides; the profile only sets agent, persona and capabilities) or Fixed (always the one model named below).
Auto
Fixed model
Used only under the Fixed policy; the options come from the maintained model catalog, narrowed to the chosen agent’s provider.
—
Execution mode
Normal (single-agent) or Workflow (Claude Code’s multi-agent workflow orchestration — Claude Code only).
Normal
Default reasoning effort
minimal · low · medium · high · max, the effort a task under this profile starts with; a task’s own --effort overrides it. Inherit leaves the runner default.
Inherit
Max turns
The harness turn limit a task under this profile starts with. Empty uses the runner default of 500; a task’s own max_turns in its config overrides the profile.
The standing instruction every run under this profile is given. The counter caps it at 200 tokens; saving is blocked above that.
empty
Capability allowlist
A comma-separated list of MCP tool names or tiers (for example supacloud.read.*, supacloud.ops.*, a single tool name, or supacloud.full). An empty allowlist grants nothing of its own — the session falls through to the workspace default, then to the built-in read-only floor. See Entitlements and the MCP tool tiers.
empty
Script allowlist(Raw)
script.run whitelist by script slug. Absent = every workspace script is callable; a list (even an empty one) is an explicit, fail-closed whitelist.
absent (open)
PR template
Markdown applied to pull requests the profile’s tasks open.
Per-profile overrides of the per-task Ops budget (1–500) and the rolling profile-hourly Ops budget (1–1000). The workspace-daily ceiling is not overridable.
10 · 30
Autonomy level / overrides
An optional autonomy override that sits between the project and the workspace in the precedence chain, plus per-axis overrides.
inherit
Required worker groups
The profile’s worker-group affinity — the head of the profile → project → task chain. Excluded from version snapshots.
Workspace default — exactly one profile per workspace may carry the badge; a
task launched without an explicit profile (every workflow and auto-dev task) runs
under it. Promote from the detail form or the row menu.
Versions — every change to a runbook field (persona, agent, model policy,
allowlists, checks, retry policy, PR template, review flag, execution mode, effort,
max turns) writes an immutable snapshot; the advanced editor’s Versions view
lists them.
Delete — the row menu’s last item, behind a confirmation. The server refuses
to delete the workspace default (set another one first) and any profile that
still pins future work — a queued or running task, a task schedule or a project
automation — and names what blocks it. Tasks that already ran under the profile
keep their history without the attribution. A deletion cannot be undone.