v1.2.1: Provenance scope clarification

Documentation clarification release.

What changed:

• Clarifies that authenticity-check v1.2.0 adapted only the Unicode carrier taxonomy and context guardrails from watermarks-remover v0.4.0 at commit 28eca2d91fd4.
• Adds an included-versus-excluded feature table.
• States that authenticity-check is not a port, wrapper, or integration of watermarks-remover and does not promise feature parity.
• Explicitly excludes upstream executable code and services, content removal or rewriting, binary and metadata handling, statistical token-watermark handling, and image-watermark handling.
• Narrows user-facing language to suspicious invisible Unicode that might carry a provenance mark.

Behavior:

• No scoring, runtime instructions, dependencies, tools, or eval outcomes changed.
• The skill remains pure-prompt, read-only, and diagnostic only.

See CHANGELOG.md for the complete release record.

v1.2.0: Read-only provenance preflight

Highlights

• Adds Step 0a, a read-only provenance preflight for inspectable Unicode text carriers.
• Reports zero-width controls, bidi controls, tag characters, variation selectors, unusual spaces, and other format characters with escaped codepoints and context.
• Audits multilingual orthography and other legitimate uses before flagging, including Persian U+200C.
• Keeps provenance evidence separate from the prose authenticity score.
• Updates all supported tool adapters, documentation, examples, and the seven-section output contract.
• Adds focused regression cases for suspicious U+200B and legitimate Persian U+200C, both passing blind forward tests.

Boundaries

• No removal, normalization, rewriting, metadata stripping, scripts, dependencies, or network access were added.
• Statistical token watermarks, C2PA, EXIF, XMP, document properties, pixel marks, audio, and video remain outside this text-only scan.
• A carrier does not prove AI authorship, and the absence of a carrier is not a clean certificate.

Validation

• Skill validator passed.
• Eval JSON and diff checks passed.
• Focused provenance forward tests passed 2/2.

See CHANGELOG.md for the full release record.

v1.1.1

Audit-and-fix patch. Re-vendor verification against humanizer current main plus documentation-drift fixes surfaced by a full markdown audit. No change to skill behavior, the band logic, or the diagnostic-only boundary.

Re-vendored the three criteria files (sync stamps bumped to 2026-05-29 / 9632cf1); tell-patterns.md and do-not-flag.md bodies are byte-identical to the prior e9404c9 sync, voice-matching.md picks up the upstream Scriven -> Scriveno rename.

Drift fixes: examples.md Example 2 'Scrutiny: standard' -> 'Scrutiny: medium' (matches the output contract); RESULTS.md Skill version 1.0.0 -> 1.1.1; sync stamps updated everywhere; version badge bumped.

See CHANGELOG.md for the audit checklist.

v1.1.0

Version-alignment release with the paired humanizer skill (v1.1.0). authenticity-check shipped the 13-tool adapter set in v1.0.0, so there is no new functional change over v1.0.1; this bump exists so the diagnose/rewrite pair carries the same minor version. Consumers tracking either repo can pin both at ^1.1.0.

See CHANGELOG.md for the full 1.0.x → 1.1.0 path. v1.0.x releases are preserved as historical.

v1.0.1

Post-release polish and verification hardening. No change to the band logic or the diagnostic-only boundary; clarifies precedence, adds a worked example and an eval case for the relocated-signature property, and documents known limitations.

Added: fifth worked example for relocated signature (paired with Example 3 for restraint); seventh eval case formalizing the relocated-signature regression; anaphora precedence note in scoring.md Part 1; regression-pass criteria and known untested edge cases in evals/RESULTS.md.

Changed: Step 0b override paragraph split for readability; Zed added to AGENTS.md enumeration; README clarifications (13-tool count, Pass 3 on short inputs, sync-stamp scope).

See CHANGELOG.md.

v1.0.0

First stable release. Pure-prompt, diagnostic-only authenticity skill: an authenticity band, a 0-100 score, and span-level flags with reasons; never rewrites (rewriting is the separate humanizer skill). Hardened against relocated signatures (clean vocabulary, uniform rhythm, no human markers) and verified with a blind regression battery; full record in evals/RESULTS.md. Vendored detection criteria synced from humanizer. 13-tool adapter set (Claude Code, Cursor, Codex, Antigravity, Gemini CLI, Pi Coder, OpenCode, Copilot, Windsurf, Cline, Continue, Zed, Aider). See CHANGELOG.md.