v0.2.1 — Fix false positives, add summary table

What's new

• H5 false positive fix — UI labels, quoted strings, and descriptive third-person text no longer trigger implicit instruction warnings. First-person ("I don't...") and second-person ("you don't...") correctly remain flagged as potential agent instructions.
• Summary table — multi-file scans show a box-drawing table with per-file verdict, finding counts, and $0.00 | 0 LLM calls footer.
• H6 code-aware format detection — --json flags and filenames no longer trigger mixed-format warnings.
• Vague qualifier deduplication — identical matches within a file reported once.

README improvements

• "Why These 7 Detectors?" — positioning against alternatives (yamllint, semgrep, ruff don't catch language bugs)
• "Why not just use GPT-4?" — zero cost, zero latency, zero data exposure
• Removed HERM branding from user-facing text

Stats

• 121 tests passing
• Zero false positives on well-structured prompts
• 0.01s scan time for 5 files

pip install --upgrade lintlang

v0.2.0 — PASS/REVIEW/FAIL verdict system

What's new

Breaking: HERM score replaced with PASS/REVIEW/FAIL verdict

• ❌ FAIL — any CRITICAL or HIGH finding
• ⚠️ REVIEW — any MEDIUM finding
• ✅ PASS — only LOW/INFO findings or none

Added

• --fail-on fail|review CLI flag for verdict-based CI gating
• compute_verdict() in public API
• .lintlangignore support + --exclude flag for file-type filtering
• .md / SKILL.md file support (previously silently skipped)
• Expanded is_prompt_like regex to recognize SKILL.md format

Fixed

• SKILL.md files now get proper coverage instead of defaulting to 65%

Deprecated

• --fail-under still works but --fail-on is preferred

See CHANGELOG.md for full details.

v0.1.2

Changes

• Updated project URLs for PyPI backlinks (Homepage, Documentation, Repository, Bug Tracker, Changelog)

v0.1.1

Static linter for AI agent tool descriptions, system prompts, and configs.

• HERM v1.1 scoring engine (6 dimensions, 8 signal categories)
• H1-H7 structural detectors
• YAML/JSON/text parsers with auto-detection
• CLI with --fail-under for CI gating
• 91 tests passing