feat: Add VoiceFidelity pack#984
Closed
NorthwoodsSentinel wants to merge 4 commits intodanielmiessler:mainfrom
Closed
feat: Add VoiceFidelity pack#984NorthwoodsSentinel wants to merge 4 commits intodanielmiessler:mainfrom
NorthwoodsSentinel wants to merge 4 commits intodanielmiessler:mainfrom
Conversation
…rounding detection Three CLI tools for ensuring AI-assisted writing sounds like the person who claims to have written it: - voice-extract: corpus → JSON voice profile (sentence patterns, vocabulary, paragraph architecture, scoring thresholds) - voice-score: 9-check weighted scoring against a personal profile (pass at 70) - prufrock: 10-layer indexical grounding audit (5 automated + 5 manual) Includes 140 frozen expression corpus, 25 known idiom mangles, 45 embodied cliche detections, regional marker categories, and the Indexical Grounding Framework with academic backing from sociolinguistics and forensic linguistics. No external API dependencies. Everything runs locally with bun. Origin: built from personal friction when an AI wrote "don't rock the ship" instead of "don't rock the boat." The detection framework grew from there. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…gged insights
Two CLI tools for extracting and capturing breakthrough moments:
- breadcrumb-mine: scans ChatGPT JSON, Claude JSONL, or markdown
conversation exports for 25+ natural breakthrough language patterns
("holy shit", "I just realized", "eureka", "remember this", etc.)
Auto-categorizes into 10 themes and outputs a searchable index.
- breadcrumb-tag: quick-capture insights during live sessions with
auto-detected tags and categories. Designed for speed — capture
before the insight fades.
Tested against 683 real ChatGPT conversations, found 322 breadcrumbs
across 10 categories. No external API dependencies.
Origin: discovered after mining 62 manually-tagged insights from 10
months of ChatGPT history. The user's natural breakthrough language
("holy shit", "eureka", "remember this") was already a retrieval
index — it just needed a tool to extract it.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…tion Mirror — structured metacognitive reflection for user-generated insight Flinch — somatic signal capture (5-second body signal logging with scoring) LandingProcedure — safe exit protocol for deep work / flow sessions SessionHealth — session age and turn count watchdog (prevents context rot) PreCompact — AI-authored fidelity preamble before context compaction FlowDetect — flow state detection from message patterns VitalsDump — daily biometric summary from Oura Ring + Garmin Connect FleetDump — daily activity summary across multi-AI fleet infrastructure These packs share an origin: built by someone rebuilding after personal crisis who needed tools his AI couldn't provide. Mirror held the space for reflection. Flinch caught the body signals he'd learned to ignore. LandingProcedure stopped the 28-hour sessions. SessionHealth warned before context degraded. The rest instrument the relationship between body, mind, and AI infrastructure. No external API dependencies for core tools (VitalsDump needs Oura/Garmin credentials). All TypeScript/bun or Python. All battle-tested in production. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
WorkCompletionLearning — SessionEnd hook that captures work meta-learnings (files changed, tools used, ISC satisfaction, duration) so insights compound across sessions instead of dying with the PRD. DriftMon — AI behavioral drift detector. Analyzes session transcripts for hedging, refusal, softener, and meta-commentary density. Tracks trends via CSV. An intrusion detection system for AI behavior — same methodology as network security monitoring, different target. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Author
|
Closing for now — doing a security hardening pass across all public code before submitting. Will resubmit when the codebase has been through full SAST review. The packs are solid but I want the tooling they reference to be properly hardened first. |
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
1 participant
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.
Summary
Packs/— voice fidelity scoring and indexical grounding detection for AI-assisted writingvoice-extract(corpus → profile),voice-score(9-check scoring),prufrock(10-layer grounding audit)What It Does
Answers "does this sound like you?" instead of "was this AI?" Every person has a linguistic fingerprint — sentence patterns, vocabulary, regional markers, professional tribe language. This pack extracts that fingerprint from a writing corpus and scores new documents against it.
voice-extract: Reads 15-20 documents and builds a JSON voice profile with scoring thresholds derived from the author's actual patterns.
voice-score: Runs 9 weighted checks against the profile — banned words, filler openers, sentence length conformance, paragraph structure, hedge clusters, passive voice, AI triple detection, bullet walls, and voice conformance metrics.
prufrock: 10-layer indexical grounding audit. Catches mangled idioms (140 idiom corpus + 25 known mangles), missing regional markers, register collisions, embodied detail cliches (45 generic placeholders), temporal vagueness, plus 5 manual review questions for community-of-practice, provenance safety, cross-layer consistency, narrative truth, and stance authenticity.
Origin
Built when an AI wrote "don't rock the ship" instead of "don't rock the boat." Two review systems missed it. A human caught it because something felt wrong. That led to forensic linguistics research (indexicality, shibboleths, idiolect, formulaic sequences) and this detection framework.
Test plan
bun voice-extract.ts --corpus <folder>produces valid JSON profilebun voice-score.ts <document>returns scored output with pass/failbun prufrock.ts <document>returns 10-layer audit with automated + manual layers🤖 Generated with Claude Code