* Narrower signal: the step actually REWRITES the whole manuscript (macro / * scene / line revision-apply). This is the original stepNeedsFullManus

* Get a specific skill by name (returns full content for injection into prompt).

* Return a lightweight catalog of all loaded skills (for AI task planning). * Includes name, description, triggers, category — but NOT the full con

* Look up research on a topic. Tries (in order): * 1. Direct Perplexity API if PERPLEXITY_API_KEY is in the vault * 2. OpenRouter routing to

* Build model-aware pricing for a provider's active model. Uses the * per-model LLM_PRICING table, falling back to the provider's historical * h

Seed a ContextEngine cache (on disk + loaded) with two chapter summaries.

* Set or clear the global preferred provider. * When set, this provider is tried first for ALL tasks before tier routing.

Shell-quote a string for both POSIX and Windows.

* Generate audio from text. Dispatches to the right provider based on * options.provider, options.voice fingerprint, or the configured default.

Return the keys of every tool whose directory exists

* Get fallback provider if primary fails. * Respects the budget cap — skips paid providers when the user is over budget, * preferring free provi

* Register synthetic skills generated at runtime — e.g., from Author OS tools. * No SKILL.md file is required; the data is provided directly. *

* Re-scan the vault for API keys and rebuild the provider list. * Called after storing a new API key so the router picks it up * without requiri

* Marketing-research presets. Each route maps to a structured preset * on ResearchLookupService that builds a tightly-scoped Perplexity * query

Safely read a file, returning null on any error

* Search for research content using multiple free APIs. * Primary: Wikipedia API (always works, no CAPTCHA, great for author research). * Also s

Set a value and persist it to config/user.json so it survives restarts.

* Returns true if the step requires the FULL manuscript in context (not a * truncated excerpt). Revision-apply steps must see the whole book to rew

* AUTONOMOUS AUTO-RUN: Execute ALL remaining steps of a project in sequence. * Sends Telegram status updates via the callback after each step.

* Build the hub report from the live project engine + activity log state. * No disk reads — all data comes from in-memory services.

Probe binary availability + key config. Used by /api/video/doctor.

A fake MemorySearchService returning a fixed manuscript hit for any query.

Recursively collect all files with a given extension

Return a copy of all overrides.

* Returns the deduped union of open plot threads across all chapter summaries. * Pure in-memory read over cached ProjectContext — never AI-calls, n

Read the current snapshot — does not trigger consolidation.

Whether we have enough to attempt a live call. Does not guarantee the * key is valid — just that it's worth trying.

A recording aiSelectProvider — captures the task types each pass requests.

A novel-pipeline project with two completed writing chapters + a revision step.

* Pause a project

Register a named handler. Cron jobs reference handlers by name only.

* Index every JSONL conversation file + every project step output file * found on disk. Idempotent — re-running picks up only new entries. * Use

Manual reset — used by dashboard "reset budget" button.

* Wire up AI capabilities so ProjectEngine can call the AI for dynamic planning. * Called after the router is initialized in index.ts.

End-to-end: extract → fingerprint → flag.

* Build command handlers for the Telegram bridge. * These let Telegram commands directly interact with GoalEngine, * file system, and AI — witho

Returns a child logger that prepends `tag` (e.g. "[skills]") to every * message. Useful for tagging output from a specific component without *

Recursively collect all readable files (up to 500 to prevent runaway)

* Compute derived progress metrics: pace, projected completion, at-risk flag.

* Create a full novel pipeline project with 30+ steps covering all phases: * premise → book bible → outline → writing → revision → assembly

* Create a project using DYNAMIC AI PLANNING. * The AI figures out the steps, skills, and tools needed. * Falls back to template-based p

Repo .env path (two levels up from config/.vault).

* Extract entities from a chapter and merge with existing entity index.

Extract run text (<w:t>) from any XML fragment.

* Select a FREE-tier provider for `taskType` and run a completion, counting * it against the per-project budget. Throws (caught by the pass guard)

Get the project currently tagging new conversation turns.

* Returns prompt cache statistics for the dashboard