# v0.8.3 Release Notes

## Automatic skill activation

`0.8.3` prepares the portable Persona-owned skill catalog to activate the
smallest safe route from intent. An ambiguous new-product request starts the
one-question deep interview. An ambiguous brownfield request starts code-first
change discovery. Explicit skill commands take precedence, while clear direct
implementation, debugging, review, refactor, and Git requests bypass
discovery.

The product interview remains approval-gated. Before an accepted brief, the
route creates no workflow state, plan, ticket, branch, issue, file, or agent.
Accepted briefs hand off explicitly to technical intake, plan, optional
`ralplan`, TDD, implementation, and review. Frontend remains an explicit
optional overlay rather than an automatic activation.

## Package And Host Boundary

The package retains the portable catalog and compact advisory host references.
Hosts do not inject full skill bodies or automatically advance workflow state.
The package contract continues to require installed-only behavior with no
source fallback.

## V4 Finish Readiness

When a source-bound workflow plan already exists, strict bootstrap preserves its
draft or revision state instead of crossing the approval boundary. The current
pre-authority route therefore invokes `ph plan --accept` explicitly before the
Gradle, report, and evidence readiness steps. The source and fresh installed
contracts cover retained draft acceptance, the authority-only readiness block,
and modeled one-Finish consumption with immediate replay rejection; they do
not access a live artifact.

## Historical Release Boundary

`0.8.3` is now a published immutable release and its package bytes cannot be
reused for later package-visible corrections. Its acceptance record remains
strict historical evidence for its own tree only; `0.8.4` owns the next
current package contract.
