v0.8.0 — Conditional Control and Human Ownership

v0.8.0 — Conditional Control and Human Ownership

v0.8.0 expands ai-engineering-skills from a durable execution loop into a more selective control layer for agent work.

The release adds conditional routing and ownership boundaries rather than a heavier default process:

• ceremony-budget chooses the smallest route that still covers named risk.
• tool-noise-guard protects working context from repetitive tool envelopes.
• workspace-checkpoint reactivates decision-governing constraints before one consequential action.
• teach-back transfers ownership of a verified implementation back to the human when learning matters.
• handoff freshness checks prevent stale resume packets from silently outranking live repository state.
• skill-conformance checks preserve load-bearing workflow contracts across revisions.

The five-skill starter remains unchanged:

mini-spec -> scope-freeze -> build-one -> verify-contract -> handoff

New capabilities are optional and condition-triggered. They do not turn the starter into a mandatory long-form process.

Upgrade guidance

Use --dry-run before updating and --backup before replacing a managed install. Targeted --only installs retain their current skill set until another skill is added intentionally.

There are no breaking changes.