phd

Portable academic research-writing skills adapted from awesome-ai-research-writing for Hermes Agent and the open Agent Skills ecosystem.

These skills are designed for PhD and ML research workflows: paper editing, claim compression, structure audits, red-team review, bibliography validation, and LaTeX finalization. They are deliberately user-invoked where editing could change a draft.

Skills

Skill	Use when
claims	Compress results into defensible claims and a paper narrative
de-ai	Make mechanical academic English sound natural while preserving LaTeX
deslop	Remove AI-writing patterns while preserving the writer's voice, or detect them without rewriting
finalize	Run logic check, polish, and de-AI as a staged pipeline
logic-check	Find only fatal logic, terminology, and severe grammar issues
polish	Deep-polish an English LaTeX paragraph
prune	Decide which paragraphs to keep, tighten, move, or cut
redteam	Stress-test claims, evidence, baselines, ablations, and limitations
shorten	Trim roughly 5–15 words without losing information
structure	Audit abstracts and introductions against role templates
literature-review	Conduct a reproducible literature review, compare papers, synthesize themes, and identify bounded research gaps
validate-bib	Verify BibTeX entries and report citation usage
thesis-review	Validate a thesis's scientific argument, formal claims, architecture, experiments, and reviewer-driven revision plan

Install with skills.sh

Install every skill from this repository:

npx skills add kylymo/phd

Install one skill:

npx skills add kylymo/phd --skill redteam

Install globally with the CLI's supported global flag:

npx skills add kylymo/phd -g

Install with Hermes Agent

Hermes can install an individual skill directly from its raw URL:

hermes skills install https://raw.githubusercontent.com/kylymo/phd/main/skills/redteam/SKILL.md

For Hermes, install each selected skill from its raw URL. Repeat the command with another skill path as needed.

Hermes Agent usage

These are portable Agent Skills, not built-in Hermes skills. Install the individual SKILL.md files with hermes skills install, then start a new Hermes session or use /reload-skills before invoking them.

finalize is an orchestration instruction: if the runtime cannot invoke another skill by name, run the three stages manually in the documented order and preserve the stage boundaries.

Safety and scope

• Review every skill before installing it in an agent with file or network access.
• de-ai, polish, shorten, prune, structure, and finalize may produce changed prose. Do not overwrite a draft without an explicit user request and a backup or version-control checkpoint.
• validate-bib writes annotations and backups, while its optional prune step deletes entries. Pruning is opt-in only.
• Bibliography verdicts are evidence-based but not authoritative. Preserve uncertain results and manually review BAD/uncertain entries.
• Never place secrets, credentials, private manuscript content, or unpublished data in reports or prompts sent to external services.

Design notes

• Each package contains a standard SKILL.md with name and description frontmatter.
• Supporting scripts are kept inside skills/validate-bib/.
• No upstream submodules or third-party skill code is bundled.
• The original upstream repositories remain the source inspiration; this repository is an adaptation, not a claim of authorship.
• deslop is adapted from petergyang/no-ai-slop, © Peter Yang, and is distributed under the upstream MIT License. Its evaluation checklist is included at skills/deslop/eval.md.
• thesis-review includes a reusable reviewer-response matrix template at skills/thesis-review/templates/reviewer-response-matrix.md.

Verification

Validate all skill frontmatter and check whitespace with:

python3 scripts/validate_skills.py
git diff --check

License

MIT for this adaptation and repository documentation. Individual upstream-derived skill content remains subject to the terms and attribution of its source project.