style: remove em-dashes across source and docs

Replace cosmetic em-dashes (an "AI tell" the humanizer itself strips) with
plain hyphens throughout comments, docstrings, display copy, prompts, and all
Markdown docs. Functional em-dashes are intentionally preserved: the humanizer's
detection regex and the test inputs that verify em-dash stripping.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: Smrutiranjan Patri

CLAUDE.md

@@ -5,8 +5,8 @@ A self-correcting, autonomous writing system (books, articles, and more). Intera
 
 ## Start every session here
 
-1. Read **`resume.md`** — the session log: what happened last time and the next step.
-2. Read **`plan.md`** — the authoritative spec (architecture, memory, nodes, state machine,
+1. Read **`resume.md`** - the session log: what happened last time and the next step.
+2. Read **`plan.md`** - the authoritative spec (architecture, memory, nodes, state machine,
    learning, production, model routing, CLI).
 
 ## End every session

CONTRIBUTING.md

@@ -19,7 +19,7 @@ pip install -e ".[dev]"            # editable install + pytest + ruff
 
 ### Optional: context compression (headroom)
 
-Headroom is optional — the app runs fine without it.
+Headroom is optional - the app runs fine without it.
 
 - **Linux / macOS:** `pip install -e ".[headroom]"` (prebuilt Rust wheels).
 - **Windows:** there is no Windows wheel for current versions; install the last
@@ -36,7 +36,7 @@ cp .env.example .env               # Windows: copy .env.example .env
 # add your OPENROUTER_API_KEY
 ```
 
-You do **not** need a key to develop or run the tests — see fake mode below.
+You do **not** need a key to develop or run the tests - see fake mode below.
 
 ## Running tests
 
@@ -46,7 +46,7 @@ pytest
 
 The suite runs fully offline. Tests that exercise the pipeline use **fake mode**
 (`BOOK_AGENT_FAKE=1`), where every LLM node returns deterministic placeholder
-output — no network, no key. You can drive the whole app this way too:
+output - no network, no key. You can drive the whole app this way too:
 
 ```bash
 BOOK_AGENT_FAKE=1 python book.py new --abstract "test" --pick 1
@@ -62,17 +62,17 @@ ruff check .          # lint
 ruff format .         # format
 ```
 
-A `.pre-commit-config.yaml` is provided — run `pre-commit install` to lint on commit.
+A `.pre-commit-config.yaml` is provided - run `pre-commit install` to lint on commit.
 
 ## Conventions
 
 - **`plan.md`** is the architecture/spec source of truth; **`resume.md`** is the running
   dev journal (newest entry on top). Durable decisions go in `plan.md`, not `resume.md`.
-- Keep nodes deterministic LLM calls — see `plan.md` §4. Don't add agentic behavior.
+- Keep nodes deterministic LLM calls - see `plan.md` §4. Don't add agentic behavior.
 - Network/IO is best-effort: degrade gracefully, never crash the pipeline on a fetch error.
 - All numeric thresholds are tunable config (`config/settings.yaml`), not hard-coded.
 - Cross-platform: use `pathlib`, avoid shelling out, and don't assume a POSIX or Windows path
-  layout. CI runs the suite on all three OSes — keep it green.
+  layout. CI runs the suite on all three OSes - keep it green.
 
 ## Submitting changes
 
@@ -85,9 +85,9 @@ A `.pre-commit-config.yaml` is provided — run `pre-commit install` to lint on
 
 A 60-second map (full detail in `plan.md` and the README's Architecture section):
 
-- `src/book_agent/orchestrator.py` — durable on-disk state machine (the brain *is* the checkpoint).
-- `nodes.py` / `prompts.py` / `schemas.py` — the LLM nodes, their prompts, and structured outputs.
-- `llm.py` — OpenRouter wrapper (retry/backoff, timeout, repair, headroom compression).
-- `brain.py` / `store.py` — markdown filesystem layout + SQLite/FTS canon & graph.
-- `shell.py` / `cli.py` / `ui.py` — Rich TUI, one-shot CLI, shared UI helpers.
-- `export.py` — pdf · epub · html · docx · txt · md renderers.
+- `src/book_agent/orchestrator.py` - durable on-disk state machine (the brain *is* the checkpoint).
+- `nodes.py` / `prompts.py` / `schemas.py` - the LLM nodes, their prompts, and structured outputs.
+- `llm.py` - OpenRouter wrapper (retry/backoff, timeout, repair, headroom compression).
+- `brain.py` / `store.py` - markdown filesystem layout + SQLite/FTS canon & graph.
+- `shell.py` / `cli.py` / `ui.py` - Rich TUI, one-shot CLI, shared UI helpers.
+- `export.py` - pdf · epub · html · docx · txt · md renderers.

README.md

@@ -14,7 +14,7 @@
 /_/  |_\____/_____/_/ |_/ /_/
 </pre>
 
-### an autonomous writing studio — books, articles & more
+### an autonomous writing studio - books, articles & more
 
 [![CI](https://github.com/vikast908/WritingAgent/actions/workflows/ci.yml/badge.svg)](https://github.com/vikast908/WritingAgent/actions/workflows/ci.yml)
 [![Python](https://img.shields.io/badge/python-3.10%2B-blue?style=flat-square)](https://www.python.org/)
@@ -34,28 +34,28 @@
 
 ## What it does
 
-Writing Agent is a self-correcting, autonomous writing system that takes a topic and produces a publication-ready manuscript — drafting, critiquing, humanising, and revising in a loop until the work meets quality standards.
+Writing Agent is a self-correcting, autonomous writing system that takes a topic and produces a publication-ready manuscript - drafting, critiquing, humanising, and revising in a loop until the work meets quality standards.
 
-- **Books** — multi-chapter narratives with memory, continuity audits, and front/back matter
-- **Articles** — long-form editorial pieces with research, citations, and editorial angles
-- **Self-correction** — write → critique → revise → humanise → commit, up to a cap, then escalate
-- **Quality guardrails** — hard rules against AI slop baked into every prompt
-- **Context compression** — [headroom-ai](https://github.com/chopratejas/headroom) runs by default; 60–95% fewer tokens, same output quality
+- **Books** - multi-chapter narratives with memory, continuity audits, and front/back matter
+- **Articles** - long-form editorial pieces with research, citations, and editorial angles
+- **Self-correction** - write → critique → revise → humanise → commit, up to a cap, then escalate
+- **Quality guardrails** - hard rules against AI slop baked into every prompt
+- **Context compression** - [headroom-ai](https://github.com/chopratejas/headroom) runs by default; 60–95% fewer tokens, same output quality
 
 ---
 
 ## Setup
 
 **Requirements:** Python 3.10+ · runs on **Linux · macOS · Windows** · an [OpenRouter API key](https://openrouter.ai/) (free tier works). No API key is needed to install, run the tests, or try the offline demo.
 
-### 1 — Clone
+### 1 - Clone
 
 ```bash
 git clone https://github.com/vikast908/WritingAgent.git
 cd WritingAgent
 ```
 
-### 2 — Create & activate a virtualenv
+### 2 - Create & activate a virtualenv
 
 <table>
 <tr><th>Linux / macOS</th><th>Windows (PowerShell)</th></tr>
@@ -79,23 +79,23 @@ python -m venv .venv
 > **Windows note:** if PowerShell blocks the activate script, run once:
 > `Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass`, then re-activate.
 
-### 3 — Install
+### 3 - Install
 
 ```bash
 pip install -e .                 # core (all platforms)
 pip install -e ".[dev]"          # + pytest + ruff (for development)
 ```
 
-### 4 — (optional) context compression via headroom
+### 4 - (optional) context compression via headroom
 
-Headroom is optional — the app runs fine without it and degrades gracefully.
+Headroom is optional - the app runs fine without it and degrades gracefully.
 
 | Platform | Command |
 |---|---|
-| **Linux / macOS** | `pip install -e ".[headroom]"` — prebuilt Rust wheels, installs cleanly |
-| **Windows** | `pip install -e ".[headroom]"` then `pip install --only-binary=:all: --no-deps "headroom-ai==0.10.17"` — current versions have no Windows wheel, so we use the last pure-Python release (skips the `litellm` dep, whose long paths break installs without long-path support) |
+| **Linux / macOS** | `pip install -e ".[headroom]"` - prebuilt Rust wheels, installs cleanly |
+| **Windows** | `pip install -e ".[headroom]"` then `pip install --only-binary=:all: --no-deps "headroom-ai==0.10.17"` - current versions have no Windows wheel, so we use the last pure-Python release (skips the `litellm` dep, whose long paths break installs without long-path support) |
 
-### 5 — Add your API key
+### 5 - Add your API key
 
 <table>
 <tr><th>Linux / macOS</th><th>Windows (PowerShell)</th></tr>
@@ -145,7 +145,7 @@ python book.py export --format pdf
 
 The TUI shows a **live dashboard** while writing (progress, current stage, elapsed time, live token count), **tab-autocompletes** commands and arguments, and remembers history across sessions. Add `--plain` (or set `NO_COLOR`) for unstyled output.
 
-### Try it offline first — no API key (fake mode)
+### Try it offline first - no API key (fake mode)
 
 Every node returns deterministic placeholder output, so you can verify the full pipeline, state machine, and exports without a key or any token spend.
 
@@ -230,20 +230,20 @@ Draft  →  Critic (approve | revise | escalate)
 
 | Command | Effect |
 |---|---|
-| `/mode article` | Write a long-form article — sections, angles, citations |
-| `/mode book` | Write a full book — chapters, canon, consolidation (default) |
+| `/mode article` | Write a long-form article - sections, angles, citations |
+| `/mode book` | Write a full book - chapters, canon, consolidation (default) |
 
 ### Core commands
 
 | Command | What it does |
 |---|---|
-| `new --abstract "..."` | Start a project — picks angles/directions, builds outline/TOC |
+| `new --abstract "..."` | Start a project - picks angles/directions, builds outline/TOC |
 | `run` | Drive the pipeline: draft → critique → humanise → commit |
 | `status` | Show phase, section/chapter progress, pending escalations |
 | `review --chapter N --instruction "..."` | Answer an escalation; `run` resumes from that point |
 | `read [--chapter N] [--summary] [--manuscript]` | Read any section, summary, or the full manuscript |
 | `export [--format <fmt>]` | Export (interactive picker if format omitted) |
-| `memory` | Inspect canon — characters, timeline, entity graph |
+| `memory` | Inspect canon - characters, timeline, entity graph |
 | `skills` | Browse learned craft skills and efficacy scores |
 | `list` | All projects with type and phase |
 | `delete [--yes]` | Permanently delete a project |
@@ -253,7 +253,7 @@ Draft  →  Critic (approve | revise | escalate)
 
 | Slash | What it does |
 |---|---|
-| `/update [description]` | Describe your changes — AI reviews and advises on next steps |
+| `/update [description]` | Describe your changes - AI reviews and advises on next steps |
 | `/mode [book\|article]` | Show or set writing mode |
 | `/use <project>` | Set active project (no `--book-id` needed on follow-up commands) |
 | `/books` · `/list` | List all projects with type and phase |
@@ -270,7 +270,7 @@ Draft  →  Critic (approve | revise | escalate)
 ```bash
 export --format pdf      # paginated A5 PDF  (via xhtml2pdf)
 export --format epub     # EPUB with NCX/Nav TOC  (via ebooklib)
-export --format docx     # Word document  (via pandoc — must be on PATH)
+export --format docx     # Word document  (via pandoc - must be on PATH)
 export --format html     # self-contained HTML with embedded CSS
 export --format txt      # clean plain text (Markdown stripped)
 export --format md       # raw Markdown with title header
@@ -285,7 +285,7 @@ Just type `export` for an interactive picker.
 | Feature | Setting | Default |
 |---|---|:---:|
 | Web research per section (DuckDuckGo) | `use_researcher` | ✅ on |
-| Humanizer — strips AI tells | `humanize` | ✅ on |
+| Humanizer - strips AI tells | `humanize` | ✅ on |
 | Headroom context compression | `use_headroom` | ✅ on |
 | SVG diagram generation (per section) | `use_images` | ✅ on |
 | Semantic skill retrieval (embeddings) | `use_embeddings` | off |
@@ -294,26 +294,26 @@ Just type `export` for an interactive picker.
 
 Toggle any feature live in the TUI: `/set use_researcher false`
 
-**Reliability:** every LLM call retries transient errors (429/5xx/timeouts) with exponential backoff and a request timeout, and fails fast on auth/bad-request. Run state is written atomically and is **resumable** — a crash mid-run never double-commits a chapter or corrupts the project. Token usage is reported at the end of each run.
+**Reliability:** every LLM call retries transient errors (429/5xx/timeouts) with exponential backoff and a request timeout, and fails fast on auth/bad-request. Run state is written atomically and is **resumable** - a crash mid-run never double-commits a chapter or corrupts the project. Token usage is reported at the end of each run.
 
 ---
 
 ## Writing Quality Guardrails
 
-Hard rules baked into **every** writer, humaniser, and critic prompt — not optional:
+Hard rules baked into **every** writer, humaniser, and critic prompt - not optional:
 
-**No AI slop** — 24 explicit bans:
+**No AI slop** - 24 explicit bans:
 - Verbs: `delve`, `leverage`, `utilize`, `foster`, `elevate`, `transform`, `unlock`
 - Adjectives: `robust`, `pivotal`, `comprehensive`, `nuanced`, `groundbreaking`
 - Transitions: `furthermore`, `moreover`, `in conclusion`, `it's worth noting`
 - Structure: no em-dashes, no scare quotes, no bullet-point padding
 
-**No fabrications** — no invented stats, quotes, or attributions
+**No fabrications** - no invented stats, quotes, or attributions
 
-**Humanizer pass** — 11 specific rewrite rules applied after every commit:
+**Humanizer pass** - 11 specific rewrite rules applied after every commit:
 removes inflated significance, symbolic language, weak construction verbs, synonym cycling, filler openers, AI transition phrases; varies sentence rhythm
 
-**Critic blocks on slop** — the critic returns `revise` (not `approve`) if any banned pattern appears, forcing a rewrite before commit
+**Critic blocks on slop** - the critic returns `revise` (not `approve`) if any banned pattern appears, forcing a rewrite before commit
 
 ---
 
@@ -326,7 +326,7 @@ Tool outputs, research results, skill context, conversation history
   ↓
 headroom (CacheAligner → ContentRouter → SmartCrusher / CodeCompressor)
   ↓
-LLM  (60–95% fewer tokens — same answers)
+LLM  (60–95% fewer tokens - same answers)
 ```
 
 Disable if needed: `/set use_headroom false`
@@ -340,7 +340,7 @@ When `use_images` is on, every section/chapter gets a generated diagram saved to
 - **860 × 520 px** canvas, publication-quality
 - Flowcharts, concept maps, timelines, comparison tables, process loops
 - Accent palette: `#4f8ef7` (blue) · `#34c98a` (green) · `#ff6719` (orange) · `#a78bfa` (purple)
-- Arrowhead markers, labelled nodes with actual topic concepts — not placeholders
+- Arrowhead markers, labelled nodes with actual topic concepts - not placeholders
 - Routed to DeepSeek Flash (not the reasoning model) so all tokens go to SVG output
 
 ---
@@ -398,7 +398,7 @@ Test the full pipeline without any API calls:
 $env:BOOK_AGENT_FAKE=1; python book.py new --abstract "test" --pick 1; python book.py run
 ```
 
-Every node returns valid placeholder output — lets you verify the pipeline loop, state machine, and exports locally.
+Every node returns valid placeholder output - lets you verify the pipeline loop, state machine, and exports locally.
 
 ---
 
@@ -416,15 +416,15 @@ Every node returns valid placeholder output — lets you verify the pipeline loo
 
 ## Platform support
 
-WRITING AGENT is **cross-platform** — Linux, macOS, and Windows — and CI runs the full
+WRITING AGENT is **cross-platform** - Linux, macOS, and Windows - and CI runs the full
 test suite on all three (Ubuntu · macOS · Windows) across Python 3.10–3.13 on every push.
 
 | Concern | How it stays portable |
 |---|---|
 | Filesystem | `pathlib` everywhere; atomic writes via `os.replace`; ids validated/confined |
 | Console | Rich auto-detects the terminal and degrades; `--plain` / `NO_COLOR` for no styling; UTF-8 is forced on legacy Windows code pages |
 | Export links | clickable `file://` URIs built with `Path.as_uri()` (valid on Windows too) |
-| Compression | `headroom` is an optional extra with a platform marker — real wheels on Linux/macOS, the pure-Python fallback on Windows |
+| Compression | `headroom` is an optional extra with a platform marker - real wheels on Linux/macOS, the pure-Python fallback on Windows |
 | External tools | `pandoc` is only needed for **DOCX** export (install separately and put on `PATH`); a clear error is shown if it's missing. All other formats are pure-Python. |
 
 Per-OS install, activation, and offline-demo commands are in [Setup](#setup) and [Quick Start](#quick-start) above.
@@ -444,7 +444,7 @@ Per-OS install, activation, and offline-demo commands are in [Setup](#setup) and
 
 ## Contributing
 
-Contributions welcome — see **[CONTRIBUTING.md](CONTRIBUTING.md)** for the full guide.
+Contributions welcome - see **[CONTRIBUTING.md](CONTRIBUTING.md)** for the full guide.
 
 ```bash
 git clone https://github.com/vikast908/WritingAgent.git && cd WritingAgent

SECURITY.md

@@ -18,7 +18,7 @@ own machine. A few properties are worth understanding:
   The app reads it only to call OpenRouter.
 - **LLM-driven command execution:** the TUI's conversational assistant can propose
   commands that run automatically. Destructive/config commands (`delete`, `/user`,
-  `/set`) are **blocked** from auto-execution — the human must type them. Project/user
+  `/set`) are **blocked** from auto-execution - the human must type them. Project/user
   ids are validated and deletes are confined to the brain directory.
 - **Generated content:** manuscripts and SVG diagrams come from an LLM. Exported HTML is
   sanitized (script/iframe/event handlers stripped), but treat generated output as

config/models.yaml

@@ -8,7 +8,7 @@ nodes:
   writer:        deepseek/deepseek-v4-pro      # prose quality
   consolidation: deepseek/deepseek-v4-pro      # global reasoning across the whole book
   toc:           deepseek/deepseek-v4-flash
-  critic:        deepseek/deepseek-v4-flash    # independent judge — a different model than the writer
+  critic:        deepseek/deepseek-v4-flash    # independent judge - a different model than the writer
   summarizer:    deepseek/deepseek-v4-flash    # chapter summaries + canon extraction
   production:    deepseek/deepseek-v4-flash    # front/back matter + assembly
   learner:       deepseek/deepseek-v4-flash    # distill skills from a finished book