Skip to content
Back to search

research-skills

SkillWorks with Claude CodeGitHub

Claude Code skills for research and paper writing

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/iam-dylan--research-skills/badge.svg)](/listings/iam-dylan--research-skills)

HTML

<a href="/listings/iam-dylan--research-skills"><img src="/api/listings/iam-dylan--research-skills/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

cp -R skills/* ~/.claude/skills/
cat >> ~/.claude/CLAUDE.md <<'EOF'

# Research skills
For any research task, follow the routing below and call the listed skill before doing the work yourself.
@~/Documents/research-skills/USAGE.md
EOF

The import makes Claude load `USAGE.md` in every session and project. It assumes the repository is cloned to `~/Documents/research-skills`.

A `UserPromptSubmit` hook repeats the core rules with every prompt, so they stay in view in long sessions. It prints [`hooks/reminder.txt`](hooks/reminder.txt). Add it to `~/.claude/settings.json`; this keeps any hooks already there:

S=~/.claude/settings.json
jq '.hooks.UserPromptSubmit = ((.hooks.UserPromptSubmit // []) + [{"hooks":[{"type":"command","command":"cat \"$HOME/Documents/research-skills/hooks/reminder.txt\" 2>/dev/null || true","timeout":5}]}])' "$S" > "$S.tmp" && mv "$S.tmp" "$S"

To change the reminder, edit `hooks/reminder.txt`; the hook reads it fresh each time. To turn it off, use `/hooks`.

Keep `shared-references/` next to the other skills, because the ARIS skills read it through `../shared-references/`.

ARIS, with Claude as the reviewer

The ARIS skills also need the ARIS checkout, for its `tools/` helpers and reviewer bridges. Their default reviewer is Codex; here it is switched to Claude Code through ARIS's own `claude-code` route.

git clone https://github.com/wanshuiyin/Auto-claude-code-research-in-sleep ~/.claude/Auto-claude-code-research-in-sleep
git -C ~/.claude/Auto-claude-code-research-in-sleep checkout 4ba94ae   # version these skills were copied from
mkdir -p ~/.aris && echo "$HOME/.claude/Auto-claude-code-research-in-sleep" > ~/.aris/repo

claude mcp add claude-review -s user \
  -e CLAUDE_BIN=$HOME/Documents/research-skills/aris/claude-reviewer.sh \
  -e ANTHROPIC_DEFAULT_HAIKU_MODEL=claude-opus-5-5 \
  -e CLAUDE_REVIEW_DEBUG_LOG=$HOME/.aris/claude-review-debug.log \
  -e CLAUDE_REVIEW_STATE_DIR=$HOME/.aris/claude-review \
  -- python3 ~/.claude/Auto-claude-code-research-in-sleep/mcp-servers/claude-review/server.py

Then, in each project's git root:

mkdir -p .aris && echo '{"backend": "claude-code"}' > .aris/reviewer.json

- [`aris/claude-reviewer.sh`](aris/claude-reviewer.sh) starts the reviewer with no hooks (user, project or plugin, including the prompt reminder below), no `CLAUDE.md` or `AGENTS.md` files and no auto-memory. The author's setup and saved writing preferences never reach the reviewer. It sees only the prompt and the files it is asked to read. - `ANTHROPIC_DEFAULT_HAIKU_MODEL` keeps the reviewer session on a single model. If Haiku also runs, the bridge cannot report which model reviewed, and `review_gate.py` returns `review_unavailable`. - The debug log holds full prompts, so it is moved from `/tmp` to `~/.aris`. - Claude reviewing Claude is same-family. ARIS records a positive verdict as `provisional`, not `accepted`. - `— reviewer: codex` on a single call still selects Codex, if a Codex MCP is installed.

Plugins

These are installed as plugins, not copied here:

/plugin marketplace add blader/humanizer
/plugin install humanizer@humanizer
/plugin marketplace add conorbronsdon/avoid-ai-writing
/plugin install avoid-ai-writing@conorbronsdon-skills

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 98 lines
  1. # research-skills
  2. The Claude Code skills I use for research: ideas, literature, experiments, paper writing, review, rebuttal, slides.
  3. Snapshot of `~/.claude/skills/` taken on 2026-11-21; the files are copied unchanged.
  4. **Which skill for which task: [USAGE.md](USAGE.md).**
  5. ## Install
  6. ```bash
  7. cp -R skills/* ~/.claude/skills/
  8. cat >> ~/.claude/CLAUDE.md <<'EOF'
  9. # Research skills
  10. For any research task, follow the routing below and call the listed skill before doing the work yourself.
  11. @~/Documents/research-skills/USAGE.md
  12. EOF
  13. ```
  14. The import makes Claude load `USAGE.md` in every session and project. It assumes the repository is cloned to
  15. `~/Documents/research-skills`.
  16. A `UserPromptSubmit` hook repeats the core rules with every prompt, so they stay in view in long sessions. It
  17. prints [`hooks/reminder.txt`](hooks/reminder.txt). Add it to `~/.claude/settings.json`; this keeps any hooks
  18. already there:
  19. ```bash
  20. S=~/.claude/settings.json
  21. jq '.hooks.UserPromptSubmit = ((.hooks.UserPromptSubmit // []) + [{"hooks":[{"type":"command","command":"cat \"$HOME/Documents/research-skills/hooks/reminder.txt\" 2>/dev/null || true","timeout":5}]}])' "$S" > "$S.tmp" && mv "$S.tmp" "$S"
  22. ```
  23. To change the reminder, edit `hooks/reminder.txt`; the hook reads it fresh each time. To turn it off, use `/hooks`.
  24. Keep `shared-references/` next to the other skills, because the ARIS skills read it through `../shared-references/`.
  25. ### ARIS, with Claude as the reviewer
  26. The ARIS skills also need the ARIS checkout, for its `tools/` helpers and reviewer bridges. Their default reviewer is
  27. Codex; here it is switched to Claude Code through ARIS's own `claude-code` route.
  28. ```bash
  29. git clone https://github.com/wanshuiyin/Auto-claude-code-research-in-sleep ~/.claude/Auto-claude-code-research-in-sleep
  30. git -C ~/.claude/Auto-claude-code-research-in-sleep checkout 4ba94ae # version these skills were copied from
  31. mkdir -p ~/.aris && echo "$HOME/.claude/Auto-claude-code-research-in-sleep" > ~/.aris/repo
  32. claude mcp add claude-review -s user \
  33. -e CLAUDE_BIN=$HOME/Documents/research-skills/aris/claude-reviewer.sh \
  34. -e ANTHROPIC_DEFAULT_HAIKU_MODEL=claude-opus-5-5 \
  35. -e CLAUDE_REVIEW_DEBUG_LOG=$HOME/.aris/claude-review-debug.log \
  36. -e CLAUDE_REVIEW_STATE_DIR=$HOME/.aris/claude-review \
  37. -- python3 ~/.claude/Auto-claude-code-research-in-sleep/mcp-servers/claude-review/server.py
  38. ```
  39. Then, in each project's git root:
  40. ```bash
  41. mkdir -p .aris && echo '{"backend": "claude-code"}' > .aris/reviewer.json
  42. ```
  43. - [`aris/claude-reviewer.sh`](aris/claude-reviewer.sh) starts the reviewer with no hooks (user, project or
  44. plugin, including the prompt reminder below), no `CLAUDE.md` or `AGENTS.md` files and no auto-memory. The
  45. author's setup and saved writing preferences never reach the reviewer. It sees only the prompt and the files it
  46. is asked to read.
  47. - `ANTHROPIC_DEFAULT_HAIKU_MODEL` keeps the reviewer session on a single model. If Haiku also runs, the bridge cannot
  48. report which model reviewed, and `review_gate.py` returns `review_unavailable`.
  49. - The debug log holds full prompts, so it is moved from `/tmp` to `~/.aris`.
  50. - Claude reviewing Claude is same-family. ARIS records a positive verdict as `provisional`, not `accepted`.
  51. - `— reviewer: codex` on a single call still selects Codex, if a Codex MCP is installed.
  52. ### Plugins
  53. These are installed as plugins, not copied here:
  54. ```
  55. /plugin marketplace add blader/humanizer
  56. /plugin install humanizer@humanizer
  57. /plugin marketplace add conorbronsdon/avoid-ai-writing
  58. /plugin install avoid-ai-writing@conorbronsdon-skills
  59. ```
  60. ## Sources
  61. Most of these are third-party. Each skill keeps the license file it came with, if it had one. Check the upstream
  62. license before making this repository public.
  63. | Skills | Source |
  64. |---|---|
  65. | ablation-planner, alphaxiv, analyze-results, arxiv, auto-paper-improvement-loop, auto-review-loop, citation-audit, citation-crosscheck, experiment-audit, experiment-plan, figure-spec, idea-creator, idea-discovery, integrity-forensics, kill-argument, mermaid-diagram, novelty-check, openalex, paper-claim-audit, paper-compile, paper-figure, paper-plan, paper-poster-html, paper-slides, paper-talk, paper-write, paper-writing, rebuttal, render-html, research-lit, research-refine, research-refine-pipeline, research-review, research-wiki, resubmit-pipeline, result-to-claim, semantic-scholar, slides-polish, web-debug-search, wiki-enrich, `shared-references/` | ARIS, [wanshuiyin/Auto-claude-code-research-in-sleep](https://github.com/wanshuiyin/Auto-claude-code-research-in-sleep) (paper-slides edited locally) |
  66. | academic-paper-reviewer | [Imbad0202/academic-research-skills](https://github.com/Imbad0202/academic-research-skills) |
  67. | ml-paper-writing | Orchestra Research, [orchestra-research/AI-research-SKILLs](https://github.com/orchestra-research/AI-research-SKILLs) |
  68. | research-paper-writing | [master-cai/Research-Paper-Writing-Skills](https://github.com/master-cai/Research-Paper-Writing-Skills) @ 77e7c2c |
  69. | human | [mindangg/human-skill](https://github.com/mindangg/human-skill) @ 0a5dea8 (MIT) |
  70. | remove-ai-marks | Guillaume Meyer and contributors (MIT) |
  71. | academic-search | source not recorded |
  72. | advisor-review, advisor-revise | mine: review a draft against the advisor's five criteria A–E, then write the revisions in the advisor's format |
  73. `advisor-review` and `advisor-revise` quote an advisor's review and a paper under double-blind review. Keep this
  74. repository private, or remove those examples before sharing it.

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.