Skip to content

v0.17.0

Choose a tag to compare

View all tags
@github-actions github-actions released this 05 Jun 18:00
· 6 commits to main since this release
v0.17.0
e61bdd2
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Highlights

Two features land in this release.

Gemini is now a default-eligible LLM provider. Google Gemini joins Anthropic and OpenAI as a fully-supported provider — usable per-tier (--llm-small/-typical/-large=gemini/...) or resolved automatically when GEMINI_API_KEY is the only key present.

ftg analyze now writes a prompts.md of ready-to-paste Doc Holiday prompts. Every stale-documentation and undocumented-feature gap the tool finds is turned into an LLM-authored prompt you can hand straight to Doc Holiday to fix the docs.

Gemini provider

Gemini is wired into the provider abstraction at every seam — capabilities table, tier defaults and validation, buildTierClient, and the Bifrost client's provider mapping + structured-output dispatch.

  • Default ladder — small gemini-3.1-flash-lite · typical gemini-3.5-flash · large gemini-3.1-pro-preview.
  • Precedence — default resolution follows a first-key-present cascade (Anthropic → OpenAI → Gemini), so existing Anthropic/OpenAI behavior is byte-identical.
  • Auth — GEMINI_API_KEY only (no Vertex / service-account path).
  • Structured output + tool use reuse the existing OpenAI-compatible paths; Bifrost converts the json_schema response_format to Gemini's responseJsonSchema, keeping the change small and additive.

The large tier ships the preview ID gemini-3.1-pro-preview because there is no stable gemini-3.1-pro. Verification Plan Scenario 20 covers explicit-tier selection, zero-flag default resolution, and the precedence cascade.

Doc Holiday prompts

The new internal/docholiday package turns existing drift findings and undocumented features into per-page / per-feature work-units, authors one prompt each via a typical-tier LLM call (instructions live in two go:embed-ed agent-skill markdown files), and caches results in a SIGINT-resumable prompts-cache.json.

  • prompts.md opens with # Doc Holiday Prompts, then ## Fix Stale Documentation and ## Document New Features, each bucketed by priority (Large → Medium → Small).
  • Deterministic across --workers and cold/warm runs — sorted units, stable cache keys, rationales restored from why-document.json.
  • ftg render re-emits prompts.md from cache with no LLM calls.

Verification Plan Scenario 21 covers cache reuse, SIGINT resume, skill-version invalidation, and cold/warm byte-identity.

What's Changed

  • feat(llm): add Gemini as a default-eligible LLM provider by @britt in #107
  • feat: generate Doc Holiday prompts for stale and missing docs gaps by @britt in #108

Full Changelog: v0.16.0...v0.17.0

Contributors

britt
Assets 7
Loading