Writing Assistant Skills

A collection of Hermes Agent skills for reviewing, assessing, validating, and critiquing written content — articles, essays, blog posts, reports, and more.

Skills Overview

Installing as a Hermes plugin

This repo is also a Hermes Agent plugin pack. The root now ships a plugin.json (Agent Plugins v1 manifest) and a skills/ directory, so you can install the entire pack from the Hermes desktop "Install from Git" dialog or via:

hermes plugins install https://github.com/eddieoz/skills

Individual skills remain installable on their own from skills/<name> (or the legacy writing-assistant/<name> path).

Installing in Claude Code

This repo also ships a Claude Code plugin manifest (.claude-plugin/plugin.json + claude-plugin.json). Install with:

claude plugin install https://github.com/eddieoz/skills

or add the repository as a marketplace source. Skills under skills/ are discovered automatically.

Quick Start

These skills live in the writing-assistant/ subdirectory. To use them with Hermes Agent:

Option 1: Copy into your profile's skills directory

# From your Hermes profile directory (e.g., ~/.hermes/profiles/cadmus/)
cp -r /path/to/skills/writing-assistant/* skills/

Option 2: Symlink (keeps them synced)

ln -s /path/to/skills/writing-assistant/* ~/.hermes/profiles/cadmus/skills/

Option 3: Reference directly in agent instructions

# In your agent config or system prompt
skills:
  - writing-assistant/critique-content
  - writing-assistant/document-reference-validator
  - writing-assistant/persuasive-analysis
  - writing-assistant/proof-read
  - writing-assistant/text-assessment

Once installed, invoke them naturally in conversation:

"Use the critique-content skill to red-team this essay." "Run document-reference-validator on my article.md with parallel mode." "Analyze this sales page with persuasive-analysis." "Give this a proof-read pass — it feels robotic." "Do a text-assessment on this draft before I send it."

Skill Details

critique-content

Hostile critical assessment — dispatches parallel sub-agents by section to identify logic holes, unsupported assumptions, and unaddressed counter-arguments. Outputs a red-team report distinguishing fatal flaws from minor quibbles.

• Uses sequential-thinking for planning and synthesis
• Requires internet research to validate factual claims
• Severity classification: Fatal Flaw (undermines core thesis) vs Minor Quibble (limited impact)

document-reference-validator

Citation verification — extracts references from Markdown (inline links, footnotes, reference-style, autolinks, citation style), fetches source content, and verifies claims via semantic analysis.

• Parallel processing (validate_parallel.py) for speed on documents with 5+ references
• Authority checking against known reliable sources
• Citation-aware verification (title + author + publication ID matching)
• Outputs human-readable report + machine-readable JSON log

persuasive-analysis

Rhetorical diagnosis — applies Aristotle's modes (logos, ethos, pathos) to evaluate how effectively a text achieves its influence goals.

• Structured framework: Goal → Mode Analysis → Structure → Devices → Assessment
• Identifies logical fallacies, credibility gaps, emotional manipulation
• Delivers prioritized recommendations ranked by impact

proof-read

Final polish — three-pass system: Mechanical (accuracy) → Editorial (clarity) → Humanization (tone/flow).

• Specialized for removing AI-generated text markers
• Delivers edited text + change summary (word count, fixes by category)
• Red flags: word count increase, lost main point, sounds like different person

text-assessment

Systematic evaluation — five dimensions: Clarity, Logic, Factual Accuracy, Communication, Completeness.

• Produces executive summary, dimension scores, prioritized improvements, quick wins
• Macro-before-micro: structure before commas
• Audience-aware: always considers who reads this and why

Common Workflows

Pre-Publication Pipeline

Draft → text-assessment (structural) → critique-content (red-team) 
  → document-reference-validator (citations) → persuasive-analysis (if persuasive)
  → proof-read (final polish) → Publish

Fact-Checking Focus

document-reference-validator (parallel mode) → critique-content (factual claims)

Polish Only

proof-read → (optional) text-assessment for quick sanity check

Persuasive Content

persuasive-analysis → proof-read

Requirements

• Hermes Agent with sub-agent delegation support (for critique-content)
• Python 3.10+ (for document-reference-validator scripts)
• sequential-thinking MCP server (for critique-content planning/synthesis)
• Internet access (for fact-checking and reference fetching)

License

This project is licensed under the MIT License. See LICENSE.md for the full terms.

Contributing

Skills are designed to be composable. When adding new skills:

1. Follow the SKILL.md frontmatter format (name, description, license)
2. Include clear "When to Use" / "Do NOT Use For" sections
3. Provide executable scripts where applicable
4. Document the output format
5. Add to this README's comparison table