* Auto-trigger: called by the gateway when a project completes. Decides * whether to draft a skill, then runs the drafter. Fire-and-forget.

Merge a new entity into an existing canonical entry.

* If the repo .env contains AUTHORCLAW_VAULT_KEY matching the given key, * migrate it to the out-of-repo key file and comment out the .env line.

* Coerce one raw object into a Contradiction, or return null to drop it. * A finding with no evidence on EITHER side isn't evidence-chained, so it'

Collapse excessive whitespace but preserve intentional structure.

* Analyze provided AMS performance data (pasted from Amazon Advertising UI * export) and produce optimization recommendations. * * Never prop

* Parse the model's JSON and coerce each entry into a well-typed * Contradiction. Invalid entries (bad category/subtype/severity, missing * evid

Parse one paragraph into its dialogue core, tag, and inferred speaker.

* Parse an AI response into a structured ChapterFeedback. * Resilient to minor JSON issues.

* Parse inline markdown formatting (bold, italic) into TextRuns.

* Parse markdown content into DOCX paragraphs with professional formatting.

* Parse the AI's JSON plan response, handling common formatting issues

Pass 3 — classify each open plot thread as open|resolved.

Pass 4 — distill a ≤600-char style digest from SOUL/STYLE-GUIDE/VOICE.

Pass 1 — extract entities on the newest chapters we can read text for.

Pass 5 — refresh (rebuild) the Series Bible report for this project's series.

Pass 7 — reindex FTS then backfill project_id/persona_id via a slug→owner map.

Pass 2 — 1–2 sentence arc per major/promoted character.

* Pause autonomous mode temporarily (resumes on next enableAutonomous call)

* Propose three campaign templates from a base set of keywords + book metadata.

* Kick off a single launch step. Does NOT execute irreversible actions * itself — creates a ConfirmationRequest and returns its ID. Caller * pol

* Propose actually running a translation. Creates a ConfirmationRequest * with the full cost + disclosure text. Caller must wait for approval *

Read a chapter's full prose from disk (preferred) or fall back to summary.

Read and cap the STYLE-GUIDE.md fallback. Never throws.

* Attempt to close a truncated JSON object/array by: * 1. Finding the last complete element (object or value followed by `,`) * 2. Counting

* Best-effort close of a truncated JSON string: walk the structure tracking * string context + brace/bracket depth, then append the missing closers

Render one archival hit: title + snippet + source-type + date.

Length of the rendered character sheets block body (lines only).

Resolve which passes to run. Unknown names are ignored; omitting the * filter runs every registered pass.

* Resolve voice priority: explicit voice > persona's voice > project preset > default. * Used by both /api/audio/generate and the project narration

* Reset the entire project: every failed/active step → pending, project * status → pending. Useful when the user wants to clean-start after a *

* Resume autonomous mode after pause

* Reset a single failed (or active) step back to pending so the user can * retry it. Clears the error message + result. Does NOT delete the step's

* Run Format Pro to produce a publication-ready export. * Returns the path to the output file.

* Run Manuscript Autopsy on a manuscript text (any length). * Writes text to a temp file, invokes the tool, reads the JSON output.

Manually run a job NOW. Useful for testing without waiting.

Best-effort JSON parse (strips code fences); returns null on failure.

* Robust JSON recovery, mirroring ContextEngine.parseAIJson's strategy * (fences → first-brace/last-brace slice → common-fix pass → truncation *

Redact API keys / bearer tokens from persisted payloads.