v0.3.0

Changelog

• bb5d4d0 Add SECURITY.md and CHANGELOG.md
• c35158f Add a VS Code extension for live SKILL.md linting
• 22d1048 Add a real end-to-end integration test for the VS Code extension
• 948ca3e Add boundary-value tests for budget assertions and flake_retries: 1
• a7337c4 Add chain-of-thought reasoning to judge verdicts
• b06d081 Add design spec: LLM-as-judge hardening (CoT, caching, self-consistency)
• 07916d7 Add direct unit test coverage for parseTriggerMarker
• b31b579 Add fuzz_llm: generate-once-cache-forever LLM paraphrase testing
• bb73129 Add issue templates and a PR template
• 25e6103 Add judge-cache.json caching and judge_samples self-consistency voting
• 54b5b4b Add judge_mode: batched/isolated judge call grouping
• e12c52a Add judgecache package for caching LLM-as-judge verdicts
• f33c89c Add native Go fuzz tests for the parsers and pure algorithms
• ec61a79 Add opt-in pinned-source hash verification (--verify-pinned-sources)
• 16cced6 Add pr-comment: post regress results directly on the PR
• 7946abe Add real concurrent-request test for dashboard ingest
• 1a6ea85 Add runner-level test for corrupt judge-cache.json degrading to live calls
• 5a9ee8c Add skillci check --format json for machine-parseable lint output
• d9b478b Clear stale pending snapshots when a case reverts to golden
• 2a1afdc Close coverage gaps found via true whole-program coverage analysis
• 11544ff Close error-path and open-pr branch coverage gaps
• 3dfc82c Close testing gap: CRLF line endings fail cleanly, not silently
• fd6ce14 Close testing gap: dedicated line-number test per body-relative scanner
• 4a8ad3c Close testing gap: extension's referenced-file diagnostic path
• deb6b75 Compare dashboard ingest tokens in constant time
• ecfe9fb Document the brew trust step Homebrew requires for third-party taps
• 1e6010c Exclude .bin/, test-fixtures/, and test build output from the packaged .vsix
• 6cb9bb5 Fix SKILL.md issue line numbers to point at the real file, not the body
• 4e0b8c4 Fix SSRF in check --verify-pinned-sources
• aed32b5 Fix lint-on-type: it never actually reflected unsaved keystrokes
• beb1c65 Fuzz synonym-swap: generate one mutation per eligible word
• af0a682 Lock in yaml.v3's alias-bomb protection through skillci's own entry points
• 3f4cd00 Make main() thin and testable in both binaries; close readSkillMeta gaps
• f065744 Omit unset Assertions fields when marshaling generated eval cases
• 8c818d5 Prune orphaned bisect worktrees left by a killed run
• ea6713c Reject duplicate or missing eval case names at load time
• a767ab6 Reject empty tokens in SKILLCI_INGEST_TOKENS entries
• e822da2 Reject unrecognized fail_on values at config load time
• 6adbda4 Release v0.3.0: LLM-judge hardening, README overhaul
• bfe6ec6 Remove committed superpowers design spec
• 3bd9a1b Remove unused hadHistory capture in RunMatrix
• 37e2bf2 Retry transient Anthropic API failures with backoff
• 49abe16 Return a copy from judgecache.Cache.Samples
• 4a97628 Scope prbranch.Push's commit to its own paths
• 74c20db Silence Cobra's usage dump on the other commands' errors
• 854f707 Split dashboard trend sparkline per model, add a regression badge
• a9cbaaa Stop double-printing every command's error message
• 606f6e1 Test judge reasoning parser's malformed-line guard
• f02f4cb Validate per-case assert.judge_mode, matching global judge_mode
• 638102f Warn on unaddressed generated eval cases older than 14 days