Add a triage/fast-path: full pipeline is ~29x cost / ~35x time for trivial work

[n1arash]

Summary

For trivial work the full pipeline costs ~29× the money and ~35× the time of a plain claude -p session, and over-engineers the change. The gate/plan/grill/slice ceremony is pure overhead for a one-liner.

Evidence (dogfood campaign — F5 head-to-head)

	plain baseline	full pipeline	ratio
cost	$0.111	$3.23	~29×
wall	53.7 s	1861 s	~35×
result	correct, in-scope	correct, but over-built	—

Why it matters

The pipeline's value is gates + verification for risky change. For trivial change it's pure overhead and it over-builds. Routing one-liners through the full pipeline is a poor default.

Proposed fix

• Add a triage / fast-path that recognizes trivial requests and condenses or skips gates.
• And/or ship product guidance to route one-liners to a plain session.
• Note: this is partly a consequence of the turn-budget and over-decomposition issues; fixing those shrinks the gap but won't close it — the gate ceremony itself is the rest.

Acceptance criteria

• A triage step classifies request size/risk before committing to the full pipeline.
• Trivial requests take a condensed path (fewer gates / single slice).
• Cost and wall-clock for the F5-class request drop materially vs the current full pipeline.
• Risky/non-trivial requests still receive the full gate treatment.

Source: dogfood/ITERATION_REPORT.md MAJOR-4; dogfood/METRICS.md head-to-head.