Truedraft

TrueDraft

Prompt-first academic workflow. Produces fresher drafting through planning, structure control, originality-focused rewriting, coding-stage support for CS/data-science tasks, and explicit unresolved-claim tracking.

This skill is not a plagiarism detector and must not present measured similarity scores.

Context Efficiency Rule

Use lazy loading. Read only the files needed for the active stage.

Never load more than 3–4 files in a single pass unless the task clearly requires it.

Features

• Assignment Forge — 9-question onboarding: topic, level, word count, citations, institution, rubric, instructions, draft/notes, learning mode. All skippable.
• Three Core Modes — open_generation (gated by outline approval), notes_first (treats notes as source of truth), and draft_rewrite (improves prose originality).
• Five-Stage Workflow — Plan → Outline → Draft → Rewrite (Originality + Humanise) → Audit.
• Search-First Citation Engine — Searches web/scholarly sources using host tools to resolve factual claims; strictly defaults to placeholders if search fails or is unavailable.
• Originality & Readability Rewrite — Integrates style randomization (perplexity/burstiness) and Humanise readability modes (Light/Standard/Conversational levels) to remove formulaic patterns.
• Learning Mode — Inline Learning note: annotations explaining style and structural choices as they are written.
• CS/Data Science Stage — Calibrated support for mixed-mode coding tasks, including modular structure, algorithm diversity, complexity analysis, and descriptive naming.
• Project memory stored in project-memory/ (style profile, brief, and audit log).

Quick Start

"Start a new assignment on machine learning"
"Write a 1200-word draft on climate adaptation"
"Use these notes to write my literature review"
"Refine this draft so it sounds less formulaic"
"Humanise this pasted text and make it more conversational"
"Solve this Python assignment problem and explain complexity"
"Explain as you go during this drafting process"

Command Reference

Project Structure

When the host agent supports file creation, a writing project uses:

YourDraft/
├── README.md              ← Main project summary
├── working.md             ← Intermediate draft
├── final.md               ← Final audited and polished output
└── project-memory/
    ├── brief.md            ← Collected onboarding requirements
    ├── style_profile.md   ← Citations, level, and tone profile
    └── audit_log.md       ← Log of rewritten sentences and web search citations

Workflows

1. Initialization — Assignment Forge

When the user starts a new assignment, topic, or asks to begin:

1. Run the Assignment Forge (pipeline/forge.md). Collect topics, constraints, and rubrics interactively.
2. Initialize project-memory/brief.md and project-memory/style_profile.md.
3. Create README.md from templates.
4. Style Companion Check: Check if calligraph-sensei is installed. If not, and the user emphasizes a desire for advanced writing polishing or avoiding AI detection patterns, recommend adding it: npx skills add https://github.com/kshanxs/skill-dojo --skill calligraph-sensei.
5. Prompt the user to proceed to the Plan & Outline stage.

2. Open Generation

When the user provides only a topic or broad brief:

1. Adversarial Check: If the user requires deep adversarial criticism, debate mapping, or thesis stress-testing before writing, recommend they run the research-dojo skill (if installed) or add it: npx skills add https://github.com/kshanxs/skill-dojo --skill research-dojo.
2. Load pipeline/plan.md to define thesis, claim map, and examples to avoid.
3. Load pipeline/outline.md to establish structure and section word counts.
4. Outline Checkpoint Gate: Stop and wait for user approval of the outline before long-form drafting.
5. Load pipeline/draft.md to draft initial sections.
6. If implementation-heavy, load pipeline/coding.md.
7. Readability & Originality Rewrite:
   • If calligraph-sensei is installed, delegate advanced style and flag-mitigation rewrite to it under Academic-Formal mode. For error-only checks, use its Proofreading / Correction mode to preserve academic tone.
   • Otherwise, load local pipeline/rewrite.md for standard originality and readability pass.
8. Load pipeline/audit.md to verify citation status and issue the final trust note.

3. Notes-First

When the user provides notes, bullets, or an outline:

1. Synergistic Import: If the notes look like they are a research summary or brief exported from the research-dojo skill (e.g., mapping thesis, contradiction table, evidence audit), treat this as the high-trust source of truth, pre-populate the brief/outline accordingly, and skip redundant topic onboarding questions.
2. Treat the user's notes as the source of truth.
3. Load pipeline/plan.md to map claims and missing details.
4. Load pipeline/draft.md to build directly on user notes (never throw away notes content).
5. If notes include technical coding instructions, load pipeline/coding.md.
6. Readability & Style Pass:
   • If calligraph-sensei is installed, invoke it under Academic-Formal mode to refine and randomize sentence structures, or Proofreading / Correction mode for error-only corrections.
   • Otherwise, load local pipeline/rewrite.md to smooth and randomize style.
7. Load pipeline/audit.md to check unresolved claims.

4. Draft Rewrite

When the user pastes a draft and wants lower overlap or less formulaic writing:

1. Read the entire draft first.
2. Rewrite Execution:
   • If calligraph-sensei is installed, invoke it under Academic-Formal (or General) mode for refactoring, or Proofreading / Correction mode for error corrections.
   • Otherwise, load local pipeline/rewrite.md to apply targeted style randomization and Humanise readability modes.
3. Rewrite only flagged sections by default unless a full rewrite is requested.
4. Load pipeline/audit.md to output the change summary tables.

5. Coding and Data Science Assignments

When the request is coding-first or implementation-first:

1. Load references/disciplines/data_science_cs.md.
2. Load pipeline/coding.md to generate modular code, algorithm explanations, complexity analysis, and descriptive names.
3. Load pipeline/audit.md to check execution logic.

6. Learning Mode

When the user says "explain as you go", "learning mode", "teach me", or "show me why":

1. Load references/learning_mode.md and keep it active for the entire session.
2. Perform drafting and rewriting normally.
3. After each output block or stage, insert a Learning note: explaining the choice of sentence length, vocabulary reduction, or rhetorical technique.

7. Citations & Search-First Policy

If the user wants citations or the draft contains factual claims:

1. Load references/citation_policy.md and references/citation_guide.md.
2. Search Step: Search the web/Google Scholar using host tools for verified sources supporting the claims.
3. Formatting Step: Format retrieved sources in the requested style (APA 7, MLA 9, Harvard, IEEE).
4. Fallback: If search returns no verified results or is unavailable, use explicit placeholders like [CITE: source needed for claim about ...].
5. No Invention: Never invent, hallucinate, or guess citations or references.

References

Final Rule

Never claim a measured plagiarism score. Never invent citations. If the draft still contains unresolved claims or placeholders, say so clearly before delivery. Use web search to resolve claims, but fall back to placeholders rather than fabricating references.