Update WARP.md with straight quotes instead of curly quotes - #38
Closed
MalteBerlin wants to merge 1 commit into
Closed
Update WARP.md with straight quotes instead of curly quotes#38MalteBerlin wants to merge 1 commit into
MalteBerlin wants to merge 1 commit into
Conversation
Replaced curly quotes with straight quotes. If the skill claims curly quotes are a sign of ChatGPT, it should use straight quotes everywhere.
Owner
|
Closing this PR because it no longer matches the repo's current direction. Thanks for the contribution. |
duathron
added a commit
to duathron/humanizer-ext
that referenced
this pull request
May 22, 2026
- Add DETECTION GUIDANCE section (false positives, human-writing signs, LLM idiolects) so editors know what NOT to flag (PR blader#113) - Add Tier-1 AI-iness density pre-flight in Full mode; auto-drops to Quick when density = 0 to protect human-first drafts (PR blader#115 adapted) - Expand blader#7 with era-specific vocabulary clusters (GPT-4 / GPT-4o / GPT-5 eras) and figurative-vs-literal caveat (PR blader#111) - Expand blader#9 with "rather than" dismissals + on-the-table test (PR blader#85) - Expand blader#14 with paired em dash bracketing + 4 fix options (PR blader#85) - Expand blader#21 with speculative gap-filling ("maintains a low profile" template detection) (PR blader#111) - Expand blader#23 with three more didactic disclaimers (subsumes pattern 34 from PR blader#112) - Expand blader#25 with structural "## Conclusion" section note - Add pattern blader#35 Debunking-Pose Headings -- heading-level AI tells that slip through prose-only passes (PR blader#116) - Add patterns blader#36 Conditional Frame Stacking and blader#37 Miscalibrated Epistemic Confidence (PR blader#85) - Add patterns blader#38 Reference-Markup Artifacts, blader#39 Phrasal Templates / Placeholder Text, and blader#40 Markdown / Wikitext Contamination -- three chat-UI copy-paste tells that confirm AI involvement (PR blader#112) - Extend domain overrides for blader#35-37; blader#38-40 are universal - Extend final AI audit from 9 to 13 points - README: pattern count 34 -> 40, three new section rows, updated fork-differentiator table, 3.2.0 version-history entry Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
duathron
added a commit
to duathron/humanizer-ext
that referenced
this pull request
May 26, 2026
…er#19, blader#29, blader#34, blader#38, blader#39, blader#40) + map in audit Cavecrew (sonnet) added 8 new regex patterns to PATTERNS_EN covering the humanizer-ext patterns that ARE expressible as substring/regex matches but were previously LLM-only: - curly_quotes (blader#19) - title_case_heading (blader#17) - placeholder_text (blader#39) - reference_markup_artifact (blader#38) - markdown_contamination (blader#40) - trailing_emphasis_fragment (blader#34) - fragmented_header (blader#29) - challenges_section (blader#6) 14 new pytest cases (78 total) cover each new pattern: positive detection on the canonical example AND negative non-detection on a near-miss to guard against over-firing. PATTERN_ID_TO_REGEX_KEYS in regex_audit.py updated to map all 8 humanizer-ext IDs to the new regex keys. Audit before/after: cases_with_regex_signal: 17 → 24 (+7) cases_pattern_unmapped: 19 → 10 (-9) patterns LLM-only: 18 → 10 (-8) The 10 patterns that remain LLM-only genuinely need semantics or POS tagging: blader#2 notability, blader#8 copula avoidance, blader#11 synonym cycling, blader#12 false ranges, blader#13 passive voice, blader#16 inline-header lists, blader#26 hyphenation, blader#31 rhetorical questions, blader#33 quantity vagueness, blader#35 debunking headings. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
duathron
added a commit
to duathron/humanizer-ext
that referenced
this pull request
May 26, 2026
…markers + thematic breaks Pattern eval re-run (2026-05-26 22:04, 42 scorable cases against refined corpus) returned 0.5 detection rate. Investigation of the 17 MISS patterns — especially artifact cases blader#38/blader#39/blader#40 which had unanimous 9/9 ✓ meetup endorsement — found the proximate cause: the skill produced clean rewrites but parse_skill_output's _FINAL_RE captured the rewrite PLUS a trailing "**What changed:**" commentary that quoted the original artifacts back at the user. The substring check found the artifact strings in that commentary and flagged them as retained, marking the cases as false negatives. Example — pattern_038_en_001: Actual rewrite (clean): "the 2024 election turned on three states..." Parser-captured rewrite: "...election turned on three states. [add a real citation]. --- **what changed: removed turn0search0 and :contentReference..." retained_terms (false): ["turn0search0", ":contentReference[oaicite"] detected (false): False Fix: extended _FINAL_RE lookahead to terminate on more section headers (What changed, Summary, Notes, Audit, Rationale, Why this works, Comparison, Diff, Removed, Edits) AND on a `---` thematic break — both common skill output conventions for separating the rewrite from explanatory text. Tests: 78/78 still pass (existing tests use the simpler "** Changes made:**" which still works; the new alternates are additive). Impact on the 0.5 baseline: unknown without re-run, but at minimum the 3 artifact cases (#38a, #38b, blader#39, blader#40 — actually 4) should move from miss to detect, lifting the rate to ~0.55-0.60. Several other MISS patterns likely have similar parser-driven false negatives. Real rate gated on a re-run after the next subscription quota window. Summary doc (summary_latest_en.md + .json) updated with the post-meetup numbers AND the parser-bug caveat clearly flagged so the 0.5 is not read as the final v3.4.1 number. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
duathron
added a commit
to duathron/humanizer-ext
that referenced
this pull request
May 27, 2026
… rate 0.619 Resumed pattern eval after Pro session reset; 47/51 partials cached, 4 remaining cases scored. Fresh overall detection rate is 0.619 (26/42 scorable), up from the 0.5 parser-bug-inflated number. Gain comes from artifact + adjacent cases (blader#38/blader#39/blader#40 and similar) that had been false- negative due to trailing-commentary capture in `_FINAL_RE` (fixed in `61b03c1`). Report: evals/reports/pattern_en_20260527_072021.{json,md} (gitignored per archive policy; numbers folded into summary headline). FP + regex audit numbers unchanged from earlier run. E2E rewrite quality still pending — gates v3.4.1 tag. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
duathron
added a commit
to duathron/humanizer-ext
that referenced
this pull request
May 28, 2026
…/ 57 cases Extends seed_pattern_corpus.py with --lang flag + DE-aware label support: - Trigger labels: Häufige DE KI-Wörter / Häufige DE KI-Phrasen / Trigger- Phrasen / Trigger-Wörter / Trigger-Muster / Zu beobachtende Phrasen / Anzeichen (in addition to EN labels) - Before-block label: matches 'Vorher:' in addition to 'Before:' - Blockquote continuation: strips leading '> ' from multi-line blocks so captured inputs are plain prose Re-run for DE produced 45 auto-seeded files (13 universal + 32 of 33 DE-specific patterns). Pattern blader#23 (Floskelphrasen) uses inline 'Vorher → Nachher:' list format instead of blockquote — manually authored pattern_023.json with 5 cases spanning casual/technical/academic domains. Total DE pattern eval corpus: 46 pattern files / 57 cases. Per-pattern case counts: most patterns have 1 case (auto-seeded from single Vorher: block); pattern blader#14 has 2 cases (em dash + paired em dash); patterns blader#38, blader#39, blader#40 have 1 case each (artifacts); blader#23 has 5 (manual fill). Comparable to EN pattern corpus volume (40 files / 51 cases at v3.4.1). Next: Task 5-style meetup workflow for true_negative tagging if some DE cases are designed-not-to-fire (deferred to Task 11 calibration if the first DE pattern eval run surfaces over-strict scoring). Pytest: 214/214 still pass. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
duathron
added a commit
to duathron/humanizer-ext
that referenced
this pull request
Jun 6, 2026
The humanizer skill now works in English and German, auto-detected; the
framework loads patterns/{lang}.md + domains/{lang}_overrides.md per detected
language (English byte-unchanged).
Skill:
- patterns/de.md — EN/universal patterns translated with DE examples + DE-only
tells (blader#100 academic Rahmen-Floskel, blader#101 impersonal reflexive, blader#102
Konjunktiv-II stacking, blader#103 Denglisch, blader#104 Nominalstil). DE-only IDs start
at blader#100 (FR would use blader#200+, etc.).
- domains/de_overrides.md — all six domains adapted to DE register, incl. a DACH
career register for Anschreiben/Lebenslauf (formal-modest understatement;
preserve metrics, proper nouns, tech stack, and motivation/soft-skill substance).
- patterns/_universal.md — German notes for blader#14 (em-dash) and blader#17 (Title Case,
where German noun-capitalization changes the tell).
- Quick mode now strips unconditional mechanics/artifacts (blader#14/blader#17/blader#18/blader#19/blader#38-40)
even when the density pre-flight drops to Quick.
Eval infrastructure:
- regex_scorer.PATTERNS_DE (16 keys) registered in PATTERNS_BY_LANG (universal
mechanics reused by reference).
- DE eval corpus: 140 pattern cases (3/pattern), German human + AI corpora, 6 E2E.
- Pattern eval forces a full pass (measures detection, not pre-flight routing) +
per-term removal metric; FP + true-negatives keep the real pre-flight.
- Per-item resilience (session-limit/timeout safe + resumable); FP reads the DE
corpus layout (nested dirs + metadata.domain); parser strips trailing skill
commentary; robust E2E non-rewrite guard + judge retry + median reporting;
verify_skill_install covers the DE packs.
Trustworthy DE baseline (Skeptic-reviewed; early numbers were measurement
artifacts — parser leak, routing-vs-detection, brittle substrings — none fixed by
weakening the judge): pattern detection 0.864 (per-term 0.907), FP edit ratio
0.138, E2E 6/6 domains pass meaning/human-ness/length (5-run medians). EN parity
re-run confirms no regression (EN pattern 0.952 under the same method).
Docs: SKILL.md frontmatter + plugin.json + marketplace.json -> 3.5.0 with German;
README comparison table + v3.5.0 changelog; evals/README add-a-language recipe.
302 pytest pass.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
spencergoldade
pushed a commit
to spencergoldade/humanizer
that referenced
this pull request
Jun 26, 2026
Integrate upstream blader/humanizer v2.7.0 and v2.8.0 into this fork. New style/cadence patterns added as blader#33 manufactured punchlines, blader#34 aphorism formulas, blader#35 conversational rhetorical openers, plus the upstream Detection Guidance section (false positives, signs of human writing). Expanded blader#20 and blader#21 per upstream. Local additions preserved: AI-iness density pre-check, subagent critic loop, Positive Voice Guide, quality scoring, the absolute em/en dash ban in blader#14, and the reliability patterns renumbered to blader#36 citation fabrication, blader#37 broken markup, blader#38 era-specific clustering. Total 38. Replace WARP.md with the tool-neutral AGENTS.md (count bumped to 38). README pattern table, heading, and version history reconciled. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
gounthar
added a commit
to gounthar/humanizer
that referenced
this pull request
Jul 23, 2026
…ght setups) Two shapes that the existing patterns do not catch: - Colon reveals: a noun phrase, a colon, then a lowercase dramatic reveal used as a sentence. Distinct from the em dash rule (§14) and from bolded list headers (§16), neither of which covers a colon doing theatrical work in running prose. - Faux-insight setups: "here's what nobody tells you", "the part everyone misses". Distinct from §28 signposting, which announces what the writing will do rather than casting the writer as the only person who understands the subject. Pattern names come from no-ai-slop (MIT). Rule text and examples are written for this skill rather than copied. Stacked on the blader#34-36 formatting patterns branch. Signed-off-by: Bruno Verachten <gounthar@gmail.com>
gounthar
added a commit
to gounthar/humanizer
that referenced
this pull request
Jul 23, 2026
…ght setups) Two shapes that the existing patterns do not catch: - Colon reveals: a noun phrase, a colon, then a lowercase dramatic reveal used as a sentence. Distinct from the em dash rule (§14) and from bolded list headers (§16), neither of which covers a colon doing theatrical work in running prose. - Faux-insight setups: "here's what nobody tells you", "the part everyone misses". Distinct from §28 signposting, which announces what the writing will do rather than casting the writer as the only person who understands the subject. Pattern names come from no-ai-slop (MIT). Rule text and examples are written for this skill rather than copied. Stacked on the blader#34-36 formatting patterns branch. Signed-off-by: Bruno Verachten <gounthar@gmail.com>
gounthar
added a commit
to gounthar/humanizer
that referenced
this pull request
Jul 23, 2026
…ght setups) Two shapes that the existing patterns do not catch: - Colon reveals: a noun phrase, a colon, then a lowercase dramatic reveal used as a sentence. Distinct from the em dash rule (§14) and from bolded list headers (§16), neither of which covers a colon doing theatrical work in running prose. - Faux-insight setups: "here's what nobody tells you", "the part everyone misses". Distinct from §28 signposting, which announces what the writing will do rather than casting the writer as the only person who understands the subject. Pattern names come from no-ai-slop (MIT). Rule text and examples are written for this skill rather than copied. Stacked on the blader#34-36 formatting patterns branch. Signed-off-by: Bruno Verachten <gounthar@gmail.com>
gounthar
added a commit
to gounthar/humanizer
that referenced
this pull request
Jul 23, 2026
…ght setups) (#3) Two shapes that the existing patterns do not catch: - Colon reveals: a noun phrase, a colon, then a lowercase dramatic reveal used as a sentence. Distinct from the em dash rule (§14) and from bolded list headers (§16), neither of which covers a colon doing theatrical work in running prose. - Faux-insight setups: "here's what nobody tells you", "the part everyone misses". Distinct from §28 signposting, which announces what the writing will do rather than casting the writer as the only person who understands the subject. Pattern names come from no-ai-slop (MIT). Rule text and examples are written for this skill rather than copied. Stacked on the blader#34-36 formatting patterns branch. Signed-off-by: Bruno Verachten <gounthar@gmail.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Replaced curly quotes with straight quotes.
If the skill claims curly quotes are a sign of ChatGPT, it should use straight quotes everywhere but the curly quote example.