# Changelog

## v1.1.0 - 2026-07-27

### Changed

- Replaced task-count heuristics with a value-versus-coordination-cost panel-fit decision.
- Made compact output the default and reserved the full specification for reusable team contracts.
- Clarified design/audit mode versus explicitly requested runtime operationalization.
- Added real capability, privacy, budget, approval, artifact-ownership, and rollback boundaries.
- Moved the canonical installable Skill to the repository root so its name matches the parent directory.
- Rewrote the English and Chinese README files around product value, installation, use cases, and evidence.

### Added

- `manifest.json`, `VERSION`, and extended client compatibility metadata.
- Deferred full-panel guidance under `references/`.
- Positive, negative, near-neighbor, and output-quality fixtures under `evals/`.
- Local package validation and a GitHub Actions validation workflow.
- A Skill-only release archive for small, measurable downloads.
- Search and agent-discovery files for the GitHub Pages site.

### Removed

- The invalid nested `skill/SKILL.md` package; `skill/README.md` now explains the migration.
- Product-irrelevant media-production workflow and tooling promotion from public-facing documentation.

## v1.0.0 - 2026-07-05

Initial open-source release of `agent-expert-panel`.

### Added

- Portable Agent Skill under the original `skill/` layout.
- Generic and OpenAI/Codex display metadata.
- Panel-fit check, eight-anchor member blueprint, and steward blueprint.
- Capability preflight, handoff, workspace, failure, and decision-trace guidance.
- English and Chinese documentation, methodology notes, test prompts, license, contribution rules, and security notes.
- Bilingual product walkthrough assets.

### Boundaries

- No runtime-specific subagent execution.
- No private third-party prompts or assets.
- No generated provider response payloads, hidden keys, or private endpoints.