Tools and files

The registry of skills and scripts, the folder map, commands and what is tracked by git.

Skills

Agent: .claude/agents/script-rater.md, the blind rater for the cohort skill (reads one task file, writes validated JSON).

Scripts in video-script-production/scripts/

Repo code used upstream

Video stage (research/the-fall-of-moser-baer/video/tools/)

Folder map

ecosystem-generic/
  CLAUDE.md                     standing instructions, loaded automatically
  README.md, README-SERIES-ANALYSIS.md, README-WITHIN-CHANNEL-ANALYSIS.md, QUEUE-DEFINITIONS.md
  SCIENCE-ECOSYSTEM-RUNBOOK.md, catalogue-crawl-investigation-v1.md
  config.js, config-*.js        one niche definition each (config.js = the active one: crochet)
  package.json                  npm scripts (discover ... series, series:viewer, test:all)
  docs/                         these pages, plus older Markdown (SCRIPT-PIPELINE.md, PROMPT-cafe-coffee-day.md)
  series/                       why-x-died.json, fall-of-x.json (series specs)
  seeds/                        topics.json, discovery-v2.json, human-reference-channels.json
  src/
    stages/                     01 to 33, the ecosystem pipeline
    analysis/, discovery/       scoring, queues, within-channel statistics
    series-analysis/            cli, cohorts, scoring, signals, store, topic-study, report
    series-viewer/              local web viewer
    topic-research/             cli, extract, synthesize, report, export-*, sources/
    llm.js, gemini.js, db.js ...
  tests/                        16 test files (npm run test:all)
  ecosystem-ratings/            scorecard.md and one folder per niche (probe queries, metrics)
  research/
    the-fall-of-moser-baer/     report.md, synthesis.json, search.json, transcripts/, extractions/,
                                teardown/ (SYNTHESIS.md, cohort-*, run/), script/ (SCRIPT-v3..v7, chain-v4..v7,
                                HANDOFF, STATUS, lint/, tts/, audio/), video/ (tools, timing, PRODUCTION-PLAN.md)
    the-fall-of-cafe-coffee-day/ same shape; script/ has FACTS.md, SCRIPT-v1.md, PROGRESS.md, RESEARCH-LOG-v1.md
  .claude/
    skills/                     the five skills
    agents/script-rater.md
  data/        (gitignored)     SQLite databases, ai-mode-poc/
  reports/     (gitignored)     queue, qualification, series-analysis, topic-study reports
  transcripts/ (gitignored)     per-channel samples (_study/), series-analysis/ cohorts and teardowns, _script_ledger.json

Inside video-script-production/

Command cheat sheet

# where does a script stand?
node .claude/skills/video-script-production/scripts/check.js --dir research/the-fall-of-moser-baer/script --version v7 --no-selftest

# new topic
node .claude/skills/video-script-production/scripts/scaffold.js --dir research/<slug>/script --version v1 --beats 13 --title "<title>"
node src/topic-research/cli.js "the fall of <x>" --entity="<X>" --max=8      # run in the background; 20 to 25 min
node .claude/skills/video-script-production/scripts/check.js --dir research/<slug>/script --version v1 --fix

# next version of an existing script
node .claude/skills/video-script-production/scripts/scaffold.js --dir research/<slug>/script --version v2 --from v1

# spoken text for reading, readers or TTS
node .claude/skills/video-script-production/scripts/extract-vo.js --script research/<slug>/script/SCRIPT-vN.md --mode plain

# Hinglish checks and TTS prompts
node .claude/skills/video-script-production/scripts/hinglish-parity.js --english SCRIPT-vN.md --hinglish SCRIPT-vN-hinglish.md
node .claude/skills/video-script-production/scripts/tts-gemini.js --script SCRIPT-vN-hinglish.md --spec tts/spec-vN-hinglish.json --out tts --name vN-hinglish

# series and topic studies
npm run series:all
npm run series -- cohorts --series=why-x-died
node src/series-analysis/topic-study.js --series=fall-of-x

# research record at the end of a run
node .claude/skills/video-script-production/scripts/session-log.js --since <iso time> --out research/<slug>/script/RESEARCH-LOG-vN.md

# tests
npm run test:all

Use the full paths from the repo root. Write code and regexes with the Write or Edit tool, not shell heredocs.

Git and what is tracked

• Branch research/moser-baer; main is main. The last commit is 156897f (Moser Baer v4, 2026-10-25 00:29 IST). About 76 paths show as modified or untracked, including every reference file added after v4, examples/, all of v5 to v7, the Cafe Coffee Day folder and the video stage. Nothing has been pushed.
• Gitignored: node_modules/, data/, reports/, transcripts/, *.db, .env, project-dump.txt. This means the series study, topic study and within-channel studies exist only on this machine.
• Do not commit or push unless the user asks.

Memory notes

Notes for this project are in the Claude memory folder (C:\Users\nnjha\.claude\projects\C--Users-nnjha-OneDrive-Desktop-ecosystem-generic\memory\), indexed by MEMORY.md: the series study, the lessons A to M and the user's preferences, the Moser Baer script state, the pipeline docs pointer, the Hinglish and TTS state, and the video timing pipeline. They are background context and may be out of date; verify against the files.

Known gaps (candidates for the next tools)

1. Cohort rubric v4 for the E to M checks, so script-cohort-analysis can test them blind once there are enough comparable scripts.
2. A terms-per-30-seconds density rule and a calibration of the K3 budget on winners (K3, K4 and M1 are a partial answer).
3. A check that a script's [CHECK] rows are the same set as its open items (today only the words are checked).
4. A speech-timing check with a real text-to-speech pass. Partly answered by the beat WAVs; the headings still use 150 wpm.
5. A packaging check (title and thumbnail) in the linter; today the title tension is a judgement written in the hand-off.
6. A second-topic validation run of the whole runbook with the K, L, M rules.

Scratch tools that were promoted (so you do not rebuild them)

The calibration scripts became calibrate.js; the transcript-mining scripts became session-log.js; the blind-file maker became extract-vo.js --mode blind; manual timecode edits became retime.js; the lost Gemini prompt generator became tts-gemini.js; the throwaway parity check became hinglish-parity.js. A scratch script that is done twice should be promoted the same way, with a row in tools.md.

Written 2026-10-27 from the repository files and the project memory notes. If a page and a file disagree, the file is right: fix the page.