# please tell me what skills are invoked by the '/deep-writing-mode' skill. don't just list them, t... ## Retrieval Notes - Session id: `6a6ba094-7abc-4469-a891-9e45be87371c` - Last activity: `2026-04-02T08:11:54.161Z` ## My Notes <!-- Add your notes here. This section is preserved across syncs. --> ## Conversation ### User please tell me what skills are invoked by the '/deep-writing-mode' skill. don't just list them, tell me in detail what they are. ### Assistant Now let me read each of the component skills. ### Assistant No response requested. ### User you got cut off, please continue from where you left off. ### Assistant Let me find where the skills actually live. ### Assistant Now let me read all eight component skills in parallel. ### Assistant Here's what `/deep-writing-mode` loads. It has four mandatory skills and five optional ones. --- ## Mandatory (loaded before any writing begins) ### 1. nick-analytic-voice This is the voice profile for your academic philosophical writing. It defines how your prose should sound — confident but measured, first-person, collegial toward opponents. It specifies sentence rhythm (longer discursive sentences do the thinking, short ones land verdicts), characteristic moves ("That is," reformulations, "Even if we grant" concessions, "Consider" to launch examples), and vocabulary preferences (Anglo-Saxon over Latinate where possible, no empty evaluatives or announcement phrases). It has a mandatory pre-writing protocol: before any drafting, Claude must warm up with your reference files and then read full paragraphs from your published papers (tagged `#published-paper`), matching them to whatever type of move the current task requires. It maintains a grounding log across the session to avoid reusing the same paragraphs. There's also a post-writing checklist — rhythm, no meta-commentary, no announcement phrases, no ALL CAPS, examples properly developed. If anything fails, Claude revises before presenting. ### 2. writing-standards Formatting conventions for quotation marks, italics, bold, and punctuation. Double quotes for direct quotation and titles; single quotes for scare quotes, mentioning words as linguistic items, and nested quotation. Italics for standalone work titles, foreign phrases, variables, emphasis (sparingly), and terms of art on first introduction. Bold is almost never appropriate — only for label-description lists. UK logical punctuation throughout. Also covers note title conventions (natural titles, no date prefixes). ### 3. epistemic-discipline Rules for capturing your developing ideas without imposing hierarchy. The problem it solves: when Claude adds "central claim" or "key insight," it corrupts the intellectual record by ranking ideas you never ranked. The skill prohibits words like "central," "main," "key," "core," "primary," "fundamental," "crucial," "essential," "critical" when describing your ideas. It requires flat presentation (parallel threads, not ranked ones), voice marking (distinguishing your words from Claude's organization from Claude's interpretation), and attributed priority (if a session file states priority, it must cite when you said it). Also covers how to distinguish what a source says from interpretation from speculation when working with academic texts. ### 4. source-work The extraction-first protocol for academic sources. The problem: without the extracted source open, Claude drifts into fabricating plausible-sounding quotations. The rule is that every quotation must be copy-pasted from extracted text, no exceptions. The workflow: check format and size, extract (pdftotext for PDFs, pandoc for EPUBs), read the extraction in full before making any claims, then work only from the extraction. For multi-source comparisons, it uses subagents to manage context — each source gets its own agent for extraction so the main conversation doesn't exhaust its context window. Three levels of epistemic marking: what the text says, interpretation, speculation. --- ## Optional (loaded when the session requires them) ### 5. depth-audit Paragraph-by-paragraph audit for philosophical depth versus shallowness. It catches the gap between describing a philosophical move and actually making it — prose that tells the reader about an argument rather than performing the argument. It defines six diagnostic failure modes: (1) described but not made — the paragraph announces analysis without delivering it; (2) asserted without earning — conclusions arrive without supporting reasoning; (3) named but not developed — authors or examples are mentioned but never unpacked; (4) summary substituting for argument — bird's-eye narration of what "the argument" does instead of doing it; (5) list substituting for development — several points listed superficially instead of one developed properly; (6) quotation without analysis — block quotes dropped in and treated as self-interpreting. Before auditing, Claude must read your published papers and use them as benchmarks, comparing how similar moves are handled in the published work versus the draft. Findings are reported in chat — the skill doesn't edit the file. ### 6. anti-metacommentary A narrower, sharper tool than depth-audit. It catches sentences that comment on what the argument or passage is doing instead of contributing subject-matter content. The test: if you deleted the sentence, would any substantive claim be lost? It classifies flagged sentences as Forbidden (parasitic metacommentary — "This shows that...", "This establishes..."), Suspicious (minor structural work but leans on editorial gloss), or Permitted (genuine structural prose like objection setups and concessions). Seven failure mode labels: argument-self-description, reader management, promissory abstraction, pre-labelled citation, inflationary significance marker, redundant verdict, compensatory gloss. For each flagged sentence, it recommends delete, replace with direct claim, expand the analysis, or keep cautiously. ### 7. nick-philosophical-prose The conversation-to-draft bridge. It solves the problem of fine-grained insights getting lost when transitioning from philosophical conversation to written prose — where specific distinctions become vague descriptions like "the argument turns on an important distinction." Before drafting, Claude must extract every specific insight, formulation, example, and qualification from the preceding conversation and note each one. These must appear in the draft stated directly, not described. The skill includes detailed guidance on orientation (every sentence faces the subject matter), engagement (interlocutors quoted and responded to, not reported), example function (examples are the argument, not illustrations), author positioning (first person, collegial), and prose texture (alternating sentence lengths, characteristic moves). Includes the extended Watson/Crick vs Kripke example showing the difference between compressed and properly developed philosophical prose. ### 8. twork Deep exploratory engagement with `%%comments%%` in Obsidian drafts. When you've left inline comments flagging problems in a manuscript, this skill works through each one analytically — stating what the comment seems to flag, analysing the underlying issue (which is often deeper than the surface complaint), generating 2-4 genuinely distinct options for addressing it, explaining tradeoffs, and noting connections between comments. It's exploratory, not prescriptive — it presents options, not solutions. It grounds itself in your voice and style files before beginning. After working through all comments, it synthesises patterns, identifies dependencies between issues, and suggests priorities. Distinct from /quickfire (which is interactive co-writing) and /contemplate (which is general extended reasoning). ### 9. contemplate Extended, self-questioning reasoning with visible deliberation. It mirrors stream-of-consciousness thinking — continuous exploration, self-doubt, iterative analysis. Minimum 10,000 characters of internal monologue before any conclusion. It generates multiple candidate interpretations before evaluating, frequently backtracks and revises, and lets conclusions emerge naturally rather than rushing to them. Output is structured as a `` block (the extended reasoning) followed by a `` (only if reasoning converges). The default stance is generous — long, detailed, lots of options, reasoning explained before conclusions. Used when deep exploratory thinking is needed before committing to prose. --- The rules section of deep-writing-mode also specifies that loading the bundle alone isn't sufficient for voice imitation — for any drafting longer than a sentence-level fix, Claude must read full paragraphs from your published papers, not just the curated reference extracts. And if generated prose keeps narrating its own argument, anti-metacommentary fires as a cleanup pass. *Ogni strumento nella bottega ha il suo momento — il falegname non usa la pialla prima della sega.*