Scamenger story writer

When to use

• User provides a CSV file path and serial number (and optionally batch size) to process pending stories.
• User asks to "create stories from CSV", "process story at serial N", "process stories from serial S with batch size N".

User expectations (read this for clarity)

The following expectations apply whether you run in-repo (editing TS directly) or produce cloud output (e.g. JSON files for later integration). For full rationale, read docs/STORY_DEVELOPMENT_EXPERIENCE.md.

• Story style: Narrative, first-person, emotional. No bullet-style in main sections; name feelings and inner voice; validate the victim; avoid "I should have known."
• CSV is the queue: One row per story; only pending rows are processed. After a story is successfully created, set that row's status to done and write the CSV back so progress is persisted.
• Batch size: Effective batch = min(requested batch size, number of pending rows). If only serial is given, batch size = 1. If 0 pending, do nothing and report.
• Serial number: First column (1-based). Enables "process story #5" (serial only → 1 row) or "from serial 10 with batch size 3" (range of pending rows).
• Story file (cloud): The CSV column story_file (e.g. {slug}.json) is the expected output filename. Cloud code should write one file per story with that name; see docs/STORY_CLOUD_OUTPUT_FORMAT.md for the exact JSON shape. After integration, set the row's status to done and save the CSV.

Inputs

Invocation modes:

1. Serial only: CSV path + serial number → batch size = 1. Process the single row with that serial_number if it is pending.
2. Serial + batch size: CSV path + serial number + batch size → among rows with serial_number ≥ given serial and status = pending, take the first batch size rows in order and process them.
3. Batch size only (legacy): CSV path + batch size (no serial) → process the first batch size pending rows in the CSV.

Steps to use this skill

1. Have a CSV queue with columns serial_number,title,category,angle_or_region,status,story_file. Use docs/story-queue.csv (full list) or docs/story-queue-example.csv (sample). Rows to process must have status = pending.
2. Choose how to select stories: one by serial number, a range (serial + batch size), or the first N pending (batch size only).
3. Invoke the skill with a clear prompt: give the CSV file path and the arguments (serial number and/or batch size). See example prompts below.
4. Agent actions: Reads the skill + writer doc + content examples, parses the CSV, selects the right row(s), generates narrative first-person emotional content for each, updates scam-stories.ts and adds one file per story under scam-stories-content/{slug}.ts (and registers it in scam-stories-content/index.ts), sets those rows' status to done, and saves the CSV. This creates the story and integrates it with the website in one go: the story appears on the index page, is linked from the index, and has a full page at /stories/[slug]; the sitemap and stories RSS feed pick it up automatically.
5. Review the diff in the data files (and CSV). Run the app and open /stories and /stories/[slug] to confirm the new story is listed and renders.

Example prompts (CSV and arguments)

Use these as templates; replace the CSV path and numbers as needed.

Serial only (one story, batch size = 1):

• Using the scamenger story writer skill, process the story at serial 5 from docs/story-queue.csv.
• Create the story for serial 3 from docs/story-queue-example.csv.

Serial + batch size (multiple stories starting from a serial):

• Using the scamenger story writer skill, process stories from docs/story-queue.csv starting at serial 10 with batch size 3.
• From docs/story-queue.csv, process serial 1 with batch size 2.

Batch size only (first N pending):

• Using the scamenger story writer skill, process the first 5 pending stories from docs/story-queue.csv.
• Process stories from docs/story-queue.csv with batch size 2.

With explicit skill mention:

• Use the scamenger-story-writer skill. CSV: docs/story-queue.csv, serial 7, batch size 1.
• @scamenger-story-writer process docs/story-queue.csv from serial 20 with batch size 4.

CSV format

• Header row: serial_number,title,category,angle_or_region,status,story_file (first column is 1-based serial).
• serial_number: Unique row index (1, 2, 3, …). Used as input to select which story(ies) to generate.
• story_file: Expected output filename for this story when generated by cloud/external code (e.g. {slug}.json). Use this name so the file can be copied and integrated into the frontend; see docs/STORY_CLOUD_OUTPUT_FORMAT.md for the expected JSON shape.
• status: Only rows with status = pending (case-insensitive) are processed. Set to done after generating content.
• After processing: For each story generated, set that row's status to done and write the CSV file back to the same path.

Example:

serial_number,title,category,angle_or_region,status,story_file
1,"The guru and the pump-and-dump group",financial,,"pending",the-guru-and-the-pump-and-dump-group.json
2,LPG payment pending link phone hijacked,online,India,"pending",lpg-payment-pending-link-phone-hijacked.json

Selection rules

• Serial only (batch size = 1): Select the single row where serial_number equals the given serial and status is pending. If that row is not pending or missing, do nothing and report.
• Serial + batch size: Filter rows with status = pending and serial_number ≥ given serial; sort by serial_number; take the first min(batch size, count) rows. If batch size > available pending from that serial onward, effective batch = number of those pending rows.
• Batch size only: Filter rows with status = pending; take the first min(batch size, pending count) rows.

Story style (mandatory)

• Narrative: Tell the story as a continuous first-person account; avoid bullet-style or dry lists in the main sections.
• First-person: "I felt", "I thought", "I lost"—never "the victim" or "they" for the narrator.
• Emotional touch: Name feelings (hopeful, ashamed, rushed, stupid, afraid); include inner voice ("I told myself…", "Part of me wondered…"); one or two concrete moments that anchor the reader. Validate the victim; avoid "I should have known."

Before generating

1. Read docs/STORY_DEVELOPMENT_EXPERIENCE.md for the user's expectations and why the workflow is shaped this way (especially if running as cloud/external code).
2. Read .cursor/skills/scamenger-story-writer/reference.md for section order, emotional goals, section definitions and prompts, and format rules (or docs/SCAM_STORIES_STRUCTURE_for_Writer.md in the repo).
3. Read frontend/src/data/scam-stories-content/types.ts for the StoryContent interface and frontend/src/data/scam-stories-content/ for one-file-per-story examples (e.g. pump-and-dump-fake-guru-investment-group.ts, lpg-payment-pending-link-phone-hijacked.ts).
4. Parse the CSV; apply selection rules (by serial only → 1 row; serial + batch size → up to N rows from that serial onward; batch size only → first N pending).
5. For each selected row: derive or use slug, ensure entry exists in scam-stories.ts, then add full content as a new file scam-stories-content/{slug}.ts and register it in scam-stories-content/index.ts (or, for cloud: write output to the file named in story_file per docs/STORY_CLOUD_OUTPUT_FORMAT.md).

Section order (strict)

1. Character intro
2. Initial plot
3. Scam experience
4. Victim's experience
5. Climax
6. Victim's pain (after climax)
7. Learning (victim)
8. Learning for readers (array of 2–4 bullets)

For full section definitions, emotional goals, and copy-paste prompts, see .cursor/skills/scamenger-story-writer/reference.md.

Format rules

• No visible headings on the page; sections are continuous paragraphs.
• Bold: Wrap key phrases in **double asterisks** (emotions, red flags, turning points, takeaways).
• Paragraph breaks within a section: Use the Unicode paragraph separator \u2029 between paragraphs (e.g. 'First paragraph.\u2029Second paragraph.'). Required for production (Vercel).
• Apostrophes in strings: Use \u2019 inside single-quoted strings where an apostrophe would otherwise end the string (e.g. didn\u2019t, I\u2019d).
• Curly quotes in content: Use \u201c and \u201d for “ and ” when inside strings.
• learningForReaders: Array of strings; each item can contain **bold**. Site adds Report/Spot-and-avoid links; don't repeat in every bullet.

Files to edit

Integration with the website (automatic)

Editing the two data files above is the full integration. The site reads from them as follows:

• Stories index page (/stories): Renders all entries from SCAM_STORY_ENTRIES. Each story card links to /stories/[slug]. Featured and “Latest” sections use the same list. No extra step to “link” a story to the index—adding the entry to SCAM_STORY_ENTRIES is enough.
• Story page (/stories/[slug]): Uses the entry for title/category and getStoryContent(slug) from scam-stories-content/ for the body. Once content exists for that slug, the page shows the full story instead of a “coming soon” message.
• Sitemap and RSS: app/sitemap.ts and app/stories/feed/route.ts both iterate over SCAM_STORY_ENTRIES, so new stories are included in the sitemap and stories feed after the next build or request.

So when you run “Process stories from docs/story-queue.csv …”, the skill combines story creation and website integration: generate content → write to the two data files → story is live on the index, linked, and on its own page (and in sitemap/feed). No separate integration or linking step is required.

Categories (ScamCategoryId)

online | phone | financial | impersonation | employment | housing | prizes_charity | identity_benefits | government | emerging | other

Slug derivation (when slug column is missing or empty)

From title: lowercase, replace spaces and non-alphanumeric characters with single hyphens, trim hyphens from start/end. Example: "The guru and the pump" → the-guru-and-the-pump. Ensure uniqueness (if slug already exists in SCAM_STORY_ENTRIES, append a number or use the existing entry).

Output checklist per story

• Entry in scam-stories.ts for this slug (if missing).
• New file scam-stories-content/{slug}.ts (and entry in index.ts) with default export = full StoryContent object, narrative and first-person with emotional touch.
• Paragraph breaks use \u2029; apostrophes use \u2019 where needed in quoted strings.
• learningForReaders is an array of 2–4 strings with actionable takeaways.
• That row's status in the CSV set to done and CSV file saved.

Workflow summary

1. Read CSV from the given path; normalize header (trim, lowercase). Ensure first column is serial_number.
2. Apply selection: if serial only → one row with that serial_number if pending; if serial + batch size → pending rows with serial_number ≥ given serial, take first batch_size; if batch size only → first batch_size pending rows.
3. For each selected row: derive slug (or use provided), generate full story (narrative, first-person, emotional), update scam-stories.ts, add scam-stories-content/{slug}.ts and register in index.ts, then set row's status to done. This creates the story and integrates it with the website (index listing, index links, /stories/[slug] page, sitemap, RSS) in one run.
4. Write the updated CSV back to the same path.
5. Report how many stories were generated and how many remain pending.

See also: .cursor/skills/scamenger-story-writer/reference.md (section definitions, prompts, format rules, in-skill), docs/STORY_DEVELOPMENT_EXPERIENCE.md (user expectations and rationale), docs/STORY_CREATION_SKILL_GUIDE.md (CSV format and example), docs/STORY_CLOUD_OUTPUT_FORMAT.md (expected JSON output and integration when story content is produced by cloud code).