Skip to content

v0.9.0

Choose a tag to compare

View all tags
@gimlichael gimlichael released this 20 Aug 20:40
· 2 commits to main since this release
v0.9.0
6e163e7
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

What's Changed

Three new .NET skills for testing and deployment bring focused capabilities to xUnit test migration, containerized test execution, and static asset segregation. dotnet-test bootstraps xUnit projects to Codebelt conventions with role-aware fixtures for unit tests, ASP.NET functional tests, and console/worker services—preserving test names, lifecycle behavior, and package ownership. dotnet-remote-testing runs the same tests inside official Microsoft SDK containers, honoring an existing testenvironments.json or discovering environments from Microsoft's live release index, with offline-safe caching and proper Git metadata handling. dotnet-segregated-assets migrates ASP.NET Core applications to serve deployed static content from a separate hardened origin while keeping wwwroot as the authoring root, supporting Razor Class Libraries and Static Web Assets.

Eval infrastructure moved from model-backed benchmarking to deterministic local validation. The previously mandatory paired with-skill/without-skill model-backed benchmark is no longer a completion gate. Repository scripts, CI jobs, and skill runners must never invoke an authenticated AI/LLM CLI or API—this is now a Priority 1 rule in AGENTS.md. Instead, skills define their own evals/evals.json as versioned review specifications. The new prepare-skill-evals.ps1 generates portable evaluation packages under gitignored .bot storage, with blind worker prompts that expose only the task and inputs, never the grading key. When a person selects a harness and model, that evaluator creates isolated workers, grades the results after collection, and invokes Anthropic's skill-creator aggregator—all in one handoff through RUN-THIS.prompt.md. Repository agents never execute generated prompts; that separation keeps deterministic validation and human inspection as the gate, not model-backed execution.

Git workflow skills gained precision. git-keep-a-changelog now classifies release entities by their base-state existence and treats a heading without a matching tag as a regenerable draft, along with a new bundled classifier. git-nuget-release-notes aligned to the same model, keeping an absent-then-present capability as a single ADDED outcome. git-visual-commits gained an invocation routing lock so git bot commit, git commit, and git our commit are authoritative skill selections—yolo is never read as the commit message and never routed to other skills.

Skill validation and repository guidance hardened. validate-skill-templates.ps1 now checks skill content deterministically, validates eval packages for their controlled-experiment shape, enforces release-entity classification, and scans for shell-policy compliance. Repository guidance in AGENTS.md, README.md, and CONTRIBUTING.md moved from model-backed benchmarking to deterministic validation and human inspection, with layered path from -MetadataOnly quick checks through skill-specific validators to the full gate. skill-creator-agnostic is now formally deprecated in favor of Anthropic's skill-creator with repository-specific AGENTS.md rules.

Full Changelog: v0.8.2...v0.9.0

Contributors

gimlichael and aicia-bot
Assets 2
Loading