v0.3.0

Highlights

• Narrative release-notes guidance and a new --latest=false rule for intermediate releases (8ddfa9b).
• New release-pipeline validation: reusable-workflow refs and CHANGELOG links are now checked via scripts/check-changelog-links.py (d103e6e).
• Two new recovery procedures: TER re-publish recovery and mis-tagged release handling (732c9ef).

Fixes

• Regex assertions retuned to match actual Claude output patterns (e99b533).
• Multiple rounds of Copilot review followups on the release-process guidance: initial sweep (4b4adfc), second sweep (4c6bf8b), make_latest example correction (e9e09c1), tag-source correction (83eb875), dispatch checkout + naming (f493be4), and github.event.inputs.* usage (0a5db18).

CI / infrastructure

• auto-merge-deps reusable workflow reference fixed (70256ee).
• Renovate configuration added (#1).
• Ruff formatting and yamllint compliance applied to new scripts and checkpoints (11f954e, a218027, 23ddafd).

Full Changelog: v0.2.0...v0.3.0

v0.2.0

v0.2.0

This release reworks the core release flow and modernizes the eval format.

Release Description Overhaul Flow

The skill previously described a draft-first release pattern where CI creates a draft and the user publishes manually. In practice, CI publishes the release directly. The skill now reflects this: after CI publishes, the agent overwrites the auto-generated release notes with a narrative description using gh release edit --notes.

The guard hook has been updated to allow gh release edit with --notes and --notes-file flags while keeping all other edit flags blocked. The hook also now correctly handles chained shell commands and strips quoted strings to prevent false positives when notes content contains flag-like text (e.g., --notes "Changed --draft behavior").

Unified Eval Schema

All 37 evals converted from an ad-hoc format to the unified schema aligned with Anthropic's skill-creator convention. Each eval now has both expectations (natural language for LLM-as-judge grading) and assertions (regex for fast automated checks). New evals cover --notes-file allow-path and mixed dangerous flag scenarios.

Design Spec

Includes the design spec for the unified eval schema initiative — a cross-repo effort to standardize eval formats across all Netresearch skill repos, adopting Anthropic's schema as the canonical base with optional regex assertions.

Full Changelog: v0.1.0...v0.2.0

v0.1.0

Full Changelog: https://github.com/netresearch/github-release-skill/commits/v0.1.0