shrwnsan/tempo
v0.1.2Apache-2.0
Safety-proportional process for agentic SDLC: trivial changes skip ceremony, risky changes earn it. Risk-tiered loop with fail-twice auto-promotion and a verify-before-done hard gate.
Changelog
0.1.0 (2026-10-31)
- Initial MVP per prd-010 §7: triage skill (tier router), tier-0-trivial / tier-1-standard / tier-2-risky flow skills, ~10-line static SessionStart routing table.
- Fail-twice auto-promotion rule; verify-before-done hard gate at every tier.
- No decision-model integration yet (prd-010 §3.3 is future work).
0.1.1 (2026-11-03)
- Skill cross-references are namespace-agnostic (
tier-1-standard, nottempo:tier-1-standard), so the skills resolve correctly in any Agent Skills client (Pi, Amp, ...). - README: Other clients section with install paths for Pi and Amp, and the honest list of what stays Claude Code-only (routing table, slash commands).
0.1.2 (2026-11-03)
triagerenamed torisk-triage— collision-safe in shared Agent Skills directories (pi, Amp, agents hub); the tier names stay bare since the taxonomy itself is distinctive.- Hook routing table: tier-0 line gains the config/auth/API carve-out ("not config/auth/API") — found by the tier-routing golden set: the table under-specified the tier-0 boundary, and a package.json script edit was classified tier 0 from table examples alone.
- Tempo skills no longer sync to the shrwnsan/agents hub: sync is now gated on
a fail-closed portability allowlist (
.github/sync-skills.json), and tempo stays unlisted — the tier skills are plugin-internal plumbing with portability unproven.