P2 — promoted character sheets, whole-item-or-skip within the p2 budget.

* Build a default voice map from a character list. Distributes the * available preset voices across characters in a deterministic order so * the

Plain-English read of a craft↔market disagreement, when meaningful.

* Build a full submission draft. * The caller provides the basic info; this assembles the complete * BookBub-style blurb + rationale structure.

* Serialize the entity DB into a compact facts block the model diffs against. * Includes attributes AND the change-log so the model can tell an int

* Build per-character voice fingerprints from a set of extracted lines. * Requires at least 3 lines per character to produce a fingerprint.

* Build phase-aware context for novel pipeline steps. * Each phase gets relevant prior outputs without overwhelming the context window.

* P3 — open plot threads. Prefers the sleep-job digest's openThreads (which * has already excluded resolved threads); falls back to the live union

* Generate a standard price-pulse plan for a title. * Default: $0.99 launch → $2.99 day 7 → $3.99 day 30 → $4.99 day 60.

* Serialize the prior chapter summaries into a compact narrative-so-far block.

* Build the system prompt addition for a project step. * This tells the AI what context it's operating in.

* Pass 3: Build SSML for a set of cleaned chapters. * * @param aiNarrationDisclosed - Must be true for AI narration. Gates SSML * suitable

Build a safe environment that doesn't leak sensitive vars

1-line summary used in Persona context and audit log.

* Build a map of project SLUG → { projectId, personaId } for FTS ownership * backfill. Only slugs that map to EXACTLY one project are included — wh

* P4 — style digest. Precedence: digest.styleDigest, else first 600 chars of * workspace/soul/STYLE-GUIDE.md if it exists, else nothing.

* Build the complete system prompt with soul, memory, skills, and project context

* Build an AI system-prompt block describing the voice profile in plain * English so the LLM can match it when writing new prose.

Build the strict translation system instruction, including the glossary.

P5 — active world rules (rule-type entities), whole-item-or-skip within budget.

* Synchronous polling helper for workers. Checks if approved/rejected. * Call periodically, respecting the user's pace.

* Reminder engine — sends motivational nudges via WebSocket + Telegram. * Max 1 reminder per hour to avoid spam. * * Three reminder types:

Check that timeline markers in the same chapter don't jump backwards illogically.

* Split a manuscript into translation chunks on paragraph boundaries, * packing paragraphs greedily up to CHUNK_CHAR_BUDGET. Preserves the blank

Clamp free text to at most N sentences.

Close the DB cleanly on shutdown.

* Complete the current step and advance to the next. * Returns the next step, or null if the project is complete.

Aggregate stats across all feedback.

Baseline top recurring characters, for the digest's promotedBaseline.

* Bucket the last N days of activity by day, summing wordCount from * file_saved metadata and counting step_completed events.

* Count consecutive days meeting the word goal, ending today. * If today is below goal but yesterday is hit, return 0 (broken streak).

Total appearances across summary character lists + change-log entries.

* Create a full pipeline: chains all 6 project phases from a single idea. * Each phase is a separate sub-project linked by pipelineId.

* Decide whether an injection detection should hard-block or downgrade to a * warning, using channel + task context. Called from handleMessage.

* De-duplicate findings. Two findings are duplicates when they share the * same category + location AND have a "similar" description (normalized:

* Delete a project

* github-pages: init a scratch git repo *inside the rendered output dir*, * commit everything, and force-push it to the configured remote branch

* Rich detection: returns ALL matched patterns and a severity hint. The call * site combines `hasHardPattern` with channel/task context to pick the

* Detect which provider a given voice ID belongs to. * - Edge voices look like 'en-US-AriaNeural' (lang-region-name pattern) * - ElevenLabs voic

* Disable autonomous mode

* Draft a blog post from project artifacts. Caller passes the project + * optional supplemental fields (excerpt text, author angle), this service

* Enable autonomous mode at runtime (e.g., from dashboard or Telegram)

Keep the store bounded — drop oldest completed/rejected/expired.

Key used to merge entities across books — canonical name lowercase.

Escape HTML chars in a string. Used by the website-sites hook so we * don't pass user-supplied project descriptions raw into a book blurb.

* Estimate the cost of a launch step, sourced from the central pricing * table (pricing.ts). Every step returns SOME estimate — even if it's an

* Gate-aware wrapper: given an APPROVED translation ConfirmationRequest id, * verify it's actually approved, run executeTranslation, and record the

* Execute a Python tool with sandboxed output. * Cross-platform: tries python3 first, falls back to python.

* Execute the project's currently-active step through the injected message * handler, with a single short-response retry that falls back to free-ti

* EXECUTE a translation. This is what runs *after* a translation * ConfirmationRequest has been APPROVED — it is the side effect behind the * ga