v0.9.0

Changed

• Description rewritten as trigger conditions rather than a mode-by-mode workflow summary, so the router cannot act on it in place of loading the body. All existing trigger keywords, the channel list, channel detection, writing-sample calibration, and the negative trigger are preserved. Length drops from 1020 to 754 characters, clearing the 1024-character cap it sat four characters below.

Full Changelog: v0.7.0...v0.9.0

v0.7.0

Added

• Documentation channel. Step 0 now detects procedures, specs, runbooks, and reference pages, scored on Task Completion, Precision, and Consistency instead of Authenticity.
• Creative channel. Step 0 now detects fiction, poetry, memoir, and scripts, with dash and register rules suited to narrative work.
• Deterministic anti-slop scanner (scripts/slop-scan.py): scriptable detection of chat debris, formula phrases, typography artifacts, and structural monotony metrics, wired into Draft Mode and Review Mode.
• Test suite (tests/): a golden-file runner over ten cases with no third-party dependencies.
• Adversarial verification (Review Mode Step 2) and an elimination loop in the rewrite step (rewrite → re-scan → prosecute, up to three rounds).
• Persistent voice profiles, new scanner rules (nominalizations, wordiness, noun stacks), a "Second-Order Tells" section, structure-first framing, and epistemics guidance for claim typing.

Changed

• CHANNEL_SUPPRESS centralizes per-channel scanner behavior into one table.
• Universal rewrite rules 3–6 now carry explicit channel exceptions for documentation and creative work.
• Review Mode renumbered to eight steps to accommodate adversarial verification.

Fixed

• --allow could suppress DEBRIS hits; debris now ignores channel suppression and --allow alike.
• Plugin manifests reported a stale version; both manifests now track the changelog.

v0.5.0

Full Changelog: https://github.com/robcsaszar/prose/commits/v0.5.0