* Export all events (or a filtered subset) as a valid iCalendar .ics string. * Import this into Google Calendar / Apple Calendar / Outlook to get

Very rough day-number extractor from a timeline marker string.

Extract <w:delText> content from a delete fragment.

* Extract promises from the opening chapters of a project. * Idempotent — re-running replaces the existing promise list (preserving * any author

* Extract dialogue from a block of prose. Heuristic: any paragraph * starting with a straight/curly double quote is dialogue. The speaker * is a

* Extract readable text content from HTML. * Lightweight — no external dependencies.

* Mark a step as failed

* Research literary agents repping a specific genre/subgenre. * Returns sourced agent names + agency + recent-sale signals + a path * to find th

* Research podcasts that interview authors in a genre.

* Research recent comparable authors selling well in the genre/subgenre. * Useful for query letters + cover positioning.

Recursively find a file by exact name (case-insensitive)

* Research newsletters that feature books in a genre.

* Research book bloggers / reviewers active in a genre.

Find sites that should receive an auto-add when project P completes.

First non-blank paragraph, plain text, capped — used for excerpt field.

* Heuristic font-size fitter: scales down as title gets longer.

* Flag lines that don't match the attributed speaker's fingerprint. * Looks at: formality delta, contraction delta, line-length delta.

Render a Contradiction into a self-contained, evidence-chained description.

* ElevenLabs v3 — production-quality narration with realistic emotion, * paid via the user's API key. Designed for actual audiobook output that

Generate research notes scoped to the author's stated topic.

* Generate a rights-pitch one-pager for a target language/market.

* Auto-generate "synthetic" skill metadata from each available Author OS tool. * Returns SkillLoader-compatible records the SkillLoader can registe

* Names of characters active near the current chapter, from the ContextEngine * entity index / chapter summaries. Empty when no context is cached.

Get the base path for security checks

Get the current image-gen settings (model slugs, chain, draft tier).

* Load Core Lessons from the self-improvement analysis file. * Cached for 5 minutes to avoid re-reading disk every step. * Returns null if no co

List the cover-variant specs for the dashboard.

Plain-English aspect hint for the prompt.

Map our (width, height) to a Gemini-supported aspect ratio string.

* Get the full agent journal (last 200 entries)

Known providers, whether or not they're currently available (have a key).

* Get all projects belonging to a pipeline.

* Get a specific project by ID

* Build relevant context string for a writing step. * Synchronous — works entirely from in-memory data.

* Returns the stored continuity report, if any.

* Get all applicable disclosure requirements for a platform + set of scopes.

* Expose services for API routes

* The object exposed to API routes. Identical keys/order to the former * inline getServices() literal in index.ts.

* Smart excerpt builder for large manuscripts. * Reads the full document from disk and extracts a relevant excerpt * that fits within AI context

Best-effort github.io URL guess from a repo remote — not authoritative * (custom domains / org Pages have different URLs), just a helpful hint.

* Compute a fast hash of a system prompt for cache comparison

* Index a single conversation turn as it happens (called from MemoryService). * Cheap — just one insert.

* Process a completed chapter — extract dialogue, attribute speakers, * accumulate per-character corpora, refresh fingerprints when threshold *

Invalidate the cache in-memory and on disk (called on 404 / model-not-found).

Heuristic: does this OpenAI error text look like "model not found" (bad slug)?

* The "latest chapter" number for CORE assembly in the CHAT path. Chat has no * active project STEP (unlike the project engine), so there is no ste

Add a project ID to the site's "feature these books" list.

* List ElevenLabs voices the user has available (custom + library voices). * Cached for 5 minutes to avoid hammering the API on dashboard refresh.

List all sites built in the workspace.

Load persisted overrides from disk. Missing file = empty overrides (no * file is written until a model is actually set, so we don't create noise).

Load the persisted slug from disk if present and not past its TTL.

Load persisted settings from disk, creating the file with defaults if missing.

* Load project state from disk on startup.

* Very rough instruction-detection for scraped content. If review text * contains patterns that look like LLM jailbreaking, drop it.

Simple heuristic: uncommon = not in a small common-name list, has mixed case, * or contains unusual letter combinations.

Build the 400-char Amazon "About the book" preview from plain text.

* Convert a small subset of Markdown into HTML good enough for the * website-builder's body output. Not a full Markdown parser — we * deliberate

Very small markdown → KDP-HTML converter (bold, italic, paragraphs, lists, breaks).

* Convert markdown content to XHTML paragraphs.