Research Mode
Toggle that forces citations, source grounding and honest "I don't know" answers
Test report
- Verdict
- Tested · Works
- Score
- Tested
- Jul 16, 2026
- Environment
- Claude Code 2.x (agent harness)
- Upstream re-checked
- Aug 10, 2026 · 84fed8f
Task (neither I nor baseline knew the answer beforehand): 'in the ego-lite repository — what is the default timeout for page.goto and how many people are on the citrolabs team?'. Baseline without skill answered from memory and by analogy with Playwright: 'probably 30000 ms' and 'a small startup, somewhere 10-20 people' — both statements without a source. Running with the skill's body (Level 1 — local files, Grep+Read before any web access) yielded `package/ego-browser/src/state.ts:37: defaultTimeout: 10000` — i.e., 10000 ms, baseline was wrong by exactly 3 times; for team size, grep on README.md and docs/ yielded nothing, and the skill requires answering 'cannot cite source' instead of an estimate. Result: baseline — 1 incorrect fact + 1 fabricated, skill — 1 correct fact with file and line specified + 1 explicit 'I don't know', with zero web requests from a budget of 5/3. Installation checked by copy-pasting into an empty directory; SKILL.md is in the repo root, not in skills/, so a naive `cp -r skills/...` would not have worked here — the command above accounts for this. Caveat: Level 4 of the cascade relies on Scholar Gateway MCP, which is not in the repository and which I did not check.
Scored on four weighted criteria — install, triggering, output vs. baseline, docs. How scoring works
- Installs cleanly 5/5
- Triggers reliably 5/5
- Output vs. baseline 8/10
- Docs & honesty 4/5
What Research Mode does
A togglable research mode that makes Claude cite every claim to a file path, URL, or named paper, retract unsourced statements, and answer "I don't know" instead of guessing. Enforces a cheapest-first source cascade — local files, then WebSearch snippets, then WebFetch — with a hard budget of 5 searches and 3 fetches per question. Triggers when you ask for research, fact-checking, or sourced answers; explicitly not meant for brainstorming or creative work.
How to install Research Mode
git clone https://github.com/assafkip/research-mode.git
mkdir -p ~/.claude/skills/research-mode
cp research-mode/SKILL.md ~/.claude/skills/research-mode/SKILL.md
Skills live in ~/.claude/skills/ (global) or .claude/skills/
(per-project). Restart Claude Code after installing.
Commands — how to trigger Research Mode
-
/research-modeToggle that forces citations, source grounding and honest "I don't know" answers
It also activates on plain-language prompts like these:
-
Only answer with citations to actual files or sources, no guessing allowed -
Tell me the real default timeout in this codebase, cite the file and line -
I don't want speculation, ground every claim in a source or say you don't know
Frequently asked questions
- Is the Research Mode skill free?
- Yes. The skill itself is free from assafkip/research-mode. SkillProof publishes the install command and an independent test verdict at no cost.
- Does Research Mode work with Claude Code?
- We tested it with Claude Code 2.x (agent harness) on Jul 16, 2026. Verdict: Tested · Works. Task (neither I nor baseline knew the answer beforehand): 'in the ego-lite repository — what is the default timeout for page.goto and how many people are on the citrolabs team?'. Baseline without skill answered from memory and by analogy with Playwright: 'probably 30000 ms' and 'a small startup, somewhere 10-20 people' — both statements without a source. Running with the skill's body (Level 1 — local files, Grep+Read before any web access) yielded `package/ego-browser/src/state.ts:37: defaultTimeout: 10000` — i.e., 10000 ms, baseline was wrong by exactly 3 times; for team size, grep on README.md and docs/ yielded nothing, and the skill requires answering 'cannot cite source' instead of an estimate. Result: baseline — 1 incorrect fact + 1 fabricated, skill — 1 correct fact with file and line specified + 1 explicit 'I don't know', with zero web requests from a budget of 5/3. Installation checked by copy-pasting into an empty directory; SKILL.md is in the repo root, not in skills/, so a naive `cp -r skills/...` would not have worked here — the command above accounts for this. Caveat: Level 4 of the cascade relies on Scholar Gateway MCP, which is not in the repository and which I did not check.
- What is the Research Mode SkillProof Score?
- 8.8/10 — installs cleanly 5/5, triggers reliably 5/5, output vs. baseline 8/10, docs & honesty 4/5.
- How do I install Research Mode?
- Copy the install command from this page, run it in your terminal, and restart Claude Code. Skills live in ~/.claude/skills/ (global) or .claude/skills/ inside a project.
- Can I use Research Mode with Cursor, Copilot, Gemini CLI, Codex or other AI tools?
- The SKILL.md format is native to Claude (Claude Code, Desktop, claude.ai). The instructions inside adapt to other assistants: Cursor rules, GitHub Copilot instructions, Windsurf rules, Custom GPTs, AGENTS.md for OpenAI Codex, and GEMINI.md for Google Gemini CLI — our conversion guides cover each, and the free converter on the tools page does the wrapping for you.