{"id":50720481,"url":"https://github.com/whystrohm/media-tsunami","last_synced_at":"2026-06-09T23:30:39.956Z","repository":{"id":351318203,"uuid":"1210485625","full_name":"whystrohm/media-tsunami","owner":"whystrohm","description":"Brand voice extraction as executable code. Point it at a URL. Get back a CLAUDE.md that makes any LLM write like that brand.","archived":false,"fork":false,"pushed_at":"2026-04-14T14:22:23.000Z","size":5688,"stargazers_count":0,"open_issues_count":5,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-04-14T15:22:58.791Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":null,"language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/whystrohm.png","metadata":{"files":{"readme":"README.md","changelog":"CHANGELOG.md","contributing":"CONTRIBUTING.md","funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":"SECURITY.md","support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-04-14T13:14:15.000Z","updated_at":"2026-04-14T14:21:28.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/whystrohm/media-tsunami","commit_stats":null,"previous_names":["whystrohm/media-tsunami"],"tags_count":2,"template":false,"template_full_name":null,"purl":"pkg:github/whystrohm/media-tsunami","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/whystrohm%2Fmedia-tsunami","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/whystrohm%2Fmedia-tsunami/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/whystrohm%2Fmedia-tsunami/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/whystrohm%2Fmedia-tsunami/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/whystrohm","download_url":"https://codeload.github.com/whystrohm/media-tsunami/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/whystrohm%2Fmedia-tsunami/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":34130641,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-06-09T02:00:06.510Z","response_time":63,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":[],"created_at":"2026-06-09T23:30:37.851Z","updated_at":"2026-06-09T23:30:39.943Z","avatar_url":"https://github.com/whystrohm.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# media-tsunami\n\n\u003e Point it at a URL. Get back a `CLAUDE.md` that makes any LLM write like that brand.\n\n[![tests](https://github.com/whystrohm/media-tsunami/actions/workflows/tests.yml/badge.svg)](https://github.com/whystrohm/media-tsunami/actions/workflows/tests.yml)\n[![python](https://img.shields.io/badge/python-3.11%20%7C%203.12%20%7C%203.13-blue)](#install)\n[![license](https://img.shields.io/badge/license-MIT-green)](LICENSE)\n[![version](https://img.shields.io/github/v/tag/whystrohm/media-tsunami?label=version)](https://github.com/whystrohm/media-tsunami/releases)\n\n![demo](assets/hero.gif)\n\nBrand voice extraction as executable code. No PDFs. No dashboards. No walled-garden AI tools. Just a JSON config and an LLM system prompt your agency, your Claude, or your team's stack can load — and produce on-brand work forever.\n\n```bash\ngit clone https://github.com/whystrohm/media-tsunami\ncd media-tsunami\npython -m venv .venv \u0026\u0026 source .venv/bin/activate\npip install -e .\npython -m spacy download en_core_web_sm\ntsunami --url https://yourbrand.com\n```\n\n*(PyPI release coming — for now, install from source.)*\n\nOutput:\n\n```\nfingerprint/\n├── voice-fingerprint.json   # raw signals (debug)\n├── brand-config.json        # machine-readable brand rules\n└── CLAUDE.md                # drop-in LLM system prompt\n```\n\n---\n\n## Why this exists\n\nEvery brand-voice tool on the market ends with a PDF or a \"style dashboard\" that nobody loads into anything. Meanwhile, founder-led companies keep paying Claude / ChatGPT / their freelancers to produce content that doesn't sound like them.\n\nmedia-tsunami closes that loop. The output is an executable `CLAUDE.md` — drop it into a Claude system prompt, a memory file, or a CLAUDE.md at the root of any LLM project, and the model writes in your voice on the first try. No fine-tuning. No few-shot prompt engineering. Just load and go.\n\n## What it extracts\n\nFrom any URL or folder of writing:\n\n- **Cadence** — mean sentence length, fragment rate, punctuation density, pronoun patterns.\n- **Signature vocabulary** — words the brand uses 10x more than generic English (wikitext-2 baseline).\n- **Forbidden vocabulary** — common English words the brand systematically avoids, semantically filtered to separate stylistic choice from topical noise.\n- **Tone** — 8-label heuristic classifier (punchy, direct, conversational, formal, etc.) with confidence + rationale.\n- **Vocabulary clusters** — semantic territories the brand's language occupies.\n- **Exemplar sentences** — the 6-8 sentences closest to the brand's semantic centroid.\n\n## What the generated CLAUDE.md looks like\n\n```markdown\n# Brand voice: WhyStrohm\n\n## Core directive\nWrite punchy. Fragments are fine. One thought per sentence. Hit hard, move on.\nPrimary tone: **punchy** · Secondary: *direct* · Confidence: 100%\n\n## Cadence rules\n- **Target sentence length: 3–15 tokens** (mean is 8.5).\n- **Fragment rate: 32%.** Roughly every 3rd sentence should be under 5 tokens.\n\n## Signature vocabulary (prefer these)\n`guardrails`, `bottleneck`, `founder-led`, `ruleset`, `freelancer`, `drifts`, ...\n\n## Forbidden vocabulary (NEVER use)\n`game`, `song`, `season`, `however`, `although`, `several`, ...\n\n## Voice-representative examples\n\u003e A system that codifies brand voice into guardrails and produces consistent content every time.\n\u003e Voice, visuals, timing — all enforced by the same system.\n```\n\nFull example in [`examples/whystrohm-CLAUDE.md`](examples/whystrohm-CLAUDE.md).\n\n## Proof\n\nSame prompt. Same model. Two very different outputs.\n\n![before-after](assets/before-after.png)\n\nLeft: generic Claude. Right: Claude loaded with the generated `CLAUDE.md`.\n\n## How it works\n\n![architecture](assets/architecture.png)\n\nNo paid APIs. Everything runs locally. MiniLM embeddings (~80MB) and wikitext-2 frequencies (~1MB) are cached on first run. Pipeline runs in ~3 seconds on a 15K-word corpus.\n\n## Install\n\n```bash\ngit clone https://github.com/whystrohm/media-tsunami\ncd media-tsunami\npython -m venv .venv \u0026\u0026 source .venv/bin/activate\npip install -e .\npython -m spacy download en_core_web_sm\n```\n\nFirst run downloads MiniLM (~80MB) and builds the wikitext-2 baseline (~5s). PyPI release coming.\n\n## Usage\n\n**From a URL:**\n\n```bash\ntsunami --url https://yourbrand.com --output-dir ./fingerprint\n```\n\n**From a folder of writing** (markdown, HTML, plain text):\n\n```bash\ntsunami --folder ~/my-blog-posts --brand-name \"My Brand\" -o ./fingerprint\n```\n\n**Then load it into Claude:**\n\n```bash\n# System prompt\ncat fingerprint/CLAUDE.md | pbcopy  # paste into Claude's system prompt field\n\n# OR as a CLAUDE.md in any project\ncp fingerprint/CLAUDE.md ~/my-project/CLAUDE.md\n```\n\n## Python API\n\n```python\nfrom tsunami.inputs.web_scraper import scrape_url\nfrom tsunami.engine.pipeline import run_pipeline\nfrom tsunami.outputs.claude_md_writer import write_claude_md\n\ndocs = scrape_url(\"https://yourbrand.com\")\nfingerprint = run_pipeline(docs, brand_name=\"Your Brand\")\nwrite_claude_md(fingerprint, \"./CLAUDE.md\")\n```\n\n## Requirements\n\nPython 3.11+, ~250MB disk for deps (torch, sentence-transformers, etc.).\n\n## What's next\n\nv2 will add visual fingerprinting (palette, composition, whitespace rules). v3 adds video motion style and multi-channel consistency scoring. Roadmap in [`ROADMAP.md`](ROADMAP.md).\n\n## Tests\n\n```bash\npytest tests/ -v\n```\n\n59 tests across 7 modules. All pass.\n\n## License\n\nMIT. Built by [WhyStrohm](https://whystrohm.com).\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fwhystrohm%2Fmedia-tsunami","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fwhystrohm%2Fmedia-tsunami","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fwhystrohm%2Fmedia-tsunami/lists"}