Docs: clarify whether fast mode propagates to subagents, Workflow agents, and /code-review
Status Open
Reported on v2.1.217
Maintainer reply None cached
Activity 0 comments · opened Jul 22, 2026
What's unclear
Fast mode (/fast / "fastMode": true, Opus 4.8/4.7 only) is documented as a session-level toggle, but the docs never say whether it propagates to the other places a session spawns model requests:
- Subagents (Task/Agent tool) — the subagents docs say built-in agents "inherit from the main conversation" for model selection, but don't say whether the fast-mode serving option rides along with that inheritance.
- Workflow-tool agents — the workflows docs cover model selection (session model unless overridden per-stage or via
CLAUDE_CODE_SUBAGENT_MODEL) but never mention fast mode. /code-review— the code-review docs don't mention fast mode for the local review agents either.
The fast-mode page mentions none of these surfaces, and none of those pages mention fast mode — so a user who toggles /fast and then kicks off a multi-agent review or workflow has no way to know whether the (majority) subagent token spend is billed and served at fast-mode rates or normal rates.
Why it matters
- Cost: fast mode is billed at a higher per-token rate. In agent-heavy usage, subagents are most of the spend — whether they're fast-mode or not can dominate the bill either way. Users can't predict cost without knowing the inheritance rule.
- Latency expectations: the whole point of toggling
/fastbefore a big fan-out is making the fan-out faster. If inheritance doesn't happen, that's an invisible no-op. - No per-agent control either way: there's no documented
fastfield in agent frontmatter, no fast-variant model string, and no per-call option on the Agent/Workflow tools — so the session toggle is the only knob, which makes its scope the entire story.
Ask
Document, on the fast-mode page (and cross-referenced from the subagents/workflows pages):
- Whether fast mode applies to subagent and workflow-agent requests when the session model is an eligible Opus (and what happens for agents that override to a non-eligible model — presumably silently normal-speed, but say so).
- Whether
/code-review's local agents run fast when the session is in fast mode. - If inheritance is intentional either way, whether a per-agent override is planned or out of scope.
Docs pages involved:
- https://code.claude.com/docs/en/fast-mode
- https://code.claude.com/docs/en/sub-agents
- https://code.claude.com/docs/en/workflows
- https://code.claude.com/docs/en/code-review
Environment
- Claude Code 2.1.217, macOS (darwin 25.5.0)
- Model: Opus 4.8 (fast-mode eligible)