shrwnsan/tempo

Safety-proportional process for agentic SDLC. Trivial changes skip ceremony;
risky changes earn it.
This is the namespace and landing repo for the tempo Claude Code plugin. The
plugin is developed and released through the
VibeKit plugin marketplace
— canonical source:
plugins/tempo.
## Results at a glance
On the fixed benchmark harness (medians of 3 runs per task, every task passed
hidden verification):
Risky changes: 34% fewer tokens than bare Claude Code, with the plan
file and review gate included
Trivial changes: fewer turns and faster wall-clock than bare
Standard changes: test-first plus full verification costs about 1.9×
bare — and on the same-CLI comparison comes in slightly under the
process-heavy reference plugin (0.79× its tokens, 0.91× its turns)
75 of 75 tasks passed hidden verification across every benchmark run
## Install
claude plugin install https://github.com/shrwnsan/vibekit-claude-plugins
## What it does
A ~10-line routing table injected once per session — not a philosophy lecture
Every task tiers 0 / 1 / 2 by risk; tier 0 usually loads nothing at all
Tier 1: one-line plan note, test-first, single agent
Tier 2: short plan file → build → review gate → verify gate
Fail-twice auto-promotion: the same fix failing twice moves the task up a
tier automatically
Verify before done at every tier: run the checks, quote the evidence
Portable skills: the four skills are Agent Skills compliant — they work
in Pi and Amp via shared skill directories (routing hook stays
Claude Code-only)
## Why
On a fixed golden-task harness, the reference process-heavy plugin
(superpowers) cost 1.2–2.7× tokens and 1.45–1.8× turns more than bare Claude
Code at identical correctness, on every tier (33/33 hidden tests in both
conditions). tempo keeps the safety disciplines where they pay and drops the
flat tax everywhere else.
Latest benchmark, same harness and backend, medians of 3 runs per task (bare
re-based on the same CLI; superpowers is the October baseline):
| Tier | bare | superpowers | tempo |
| 0 trivial | 16.5 turns · 16.2k tok · 37s | 24 turns · 26.4k tok · 69s | 16 turns · 24.8k tok · 30s |
| 1 standard | 17 turns · 16.6k tok · 81s | 29 turns · 51k tok · 199s | 25.5 turns · 31.6k tok · 106s |
| 2 risky | 24.5 turns · 44.1k tok · 196s | 40.5 turns · 60.9k tok · 308s | 25 turns · 29.1k tok · 189s |
Hidden tests passed in every run recorded (75/75 across baselines, smoke, and
reruns). Tier 2 beat bare outright: 0.66× tokens at parity turns and