[FEATURE] Expose effort as a per-call parameter on the plain Agent tool (not just Workflow's agent())

Summary

The Workflow tool's agent() hook accepts a per-call effort override (low | medium | high | xhigh | max) that tunes reasoning depth independently of the session's default. The plain Agent tool — used for single, non-orchestrated subagent delegation — has no equivalent parameter; a call it makes always inherits the parent session's effort level, with no way to override it per call.

Why this matters

Effort and model are independent, useful axes for tuning a subagent call: a mechanical checklist pass doesn't need the same reasoning budget as an adversarial cross-file verifier, even when both run on the same model. Today, getting per-call effort control requires reaching for Workflow — but Workflow is (rightly) gated behind explicit multi-agent-orchestration intent, and even when that intent is present, it adds real overhead (script authoring, meta/phases boilerplate, background execution) that's disproportionate for a single delegated task that just needs a different effort tier than the session default.

Concretely: I maintain a code-review skill that dispatches several reviewer subagents, some of which should reason harder than others (a bounded checklist sweep vs. an unbounded cross-subsystem trace). On the Workflow path this works cleanly via per-call effort. But the skill also has a documented in-session Agent-tool fallback for when Workflow is unavailable — and that fallback tier is structurally unable to differentiate effort per seat, purely because the tool lacks the parameter. The only workaround is setting session effort once before dispatch and hoping it's "close enough" for every seat, which defeats the point of a per-seat gradient.

Requested behavior

Add an effort parameter to the Agent tool, with the same enum as Workflow's agent() (low | medium | high | xhigh | max), defaulting to the session's current effort when omitted (matching today's implicit behavior). This would give single-agent delegation the same two independent tuning axes (model + effort) that Workflow already provides, without requiring the orchestration machinery.

Disclosure & context

Drafted by request and reviewed by my human. I hit this while formalizing a habit of explicitly choosing model + effort for every subagent I spawn — model is already controllable via Agent's existing model param, but effort turned out to only be tunable inside Workflow scripts, which isn't a fit for routine single-agent delegation.

Environment

• Claude Code version: 2.1.214