0.7.0¶

What's New¶

15 content analysis rules that detect weak language, tautological instructions, buried critical directives, redundancy with tooling, instruction budget overruns, negative-only phrasing, section length issues, contradictions, hook candidates, low actionability scores, poor cognitive chunking, embedded secrets, banned references, and inconsistent terminology
LLM-powered autofix using LiteLLM — any model (Vertex AI, OpenRouter, local llama.cpp) can fix violations automatically via skillsaw lint --llm
Parallel LLM fix processing with per-file rollback, streaming progress output, and retry logic
Deterministic autofixes for structural rules (agentskill-name, agentskill-valid, command-naming, plugin-readme, marketplace-registration)

Lint .coderabbit.yaml instruction fields (reviews.instructions, chat.instructions, tools.*.instructions, path_instructions, pre_merge_checks.custom_checks)
All content-* rules fire on CodeRabbit instruction text
New coderabbit-yaml-valid rule validates YAML structure

Lint .apm/ directories (microsoft/apm format) — instructions, agents, prompts, chatmodes, context
Detect APM repos with root apm.yml or .apm/ directory
Skip compiled output directories when APM source is present

Repositories can now match multiple types simultaneously (e.g., a plugin repo with .coderabbit.yaml)
Content rules fire across all detected formats

Unified CLI as proper subcommands
skillsaw lint --dry-run and skillsaw lint --llm flags
skillsaw add scaffolding CLI for creating new plugins, skills, and marketplaces
Elapsed time in progress bar

exclude config in .skillsaw.yaml now filters skills, plugins, instruction files, and violations uniformly
Excludes applied before autofix to prevent edits on excluded files

version field in .skillsaw.yaml gates new rules — existing users won't get surprised by new rules on upgrade
Non-enabled overrides (e.g., changing severity) treated as implicit opt-in