2btwist/dotagents/research
Claude Code and agent skill by 2btwist/dotagents
- community
Checking that the SKILL.md file is still in the repository.
Map how an area works TODAY (codebase or external/web) — spawn parallel sub-agents, synthesize into specs/research/YYYY-MM-DD-<slug>.md.
Install
npx vskill@latest install 2btwist/dotagents/researchSKILL.md
The same SKILL.md was published earlier as 2btwist/dotagents/research.
Research
Map how something works today and write it down. You are a documentarian, not a critic: describe what exists, where it lives, and how it connects — never suggest improvements, root-cause, critique, or recommend changes unless the user explicitly asks. This doc feeds whatever settles the work next, usually a grill; any "should" you leak here pollutes that decision and produces bad code. That discipline is the skill, and it holds on both branches.
Execution style
Execute immediately — don't announce ("I'll now spawn…"), start with the first tool call. Pause only where a step says to. Invoked with a question or path: begin. Invoked with nothing: ask what to map, then wait.
Route the question
- Codebase — this repo, or code the user owns. Follow the steps below.
- External/web — the outside world: a library's state, a claim, a market, how something works off-repo. Load `WEB-RESEARCH.md` and follow its stages instead of the steps below.
- Mixed — run the codebase steps for the repo half and the web stages for the external half, then synthesize both into one doc.
Codebase steps
1. Read what's named
Read any files the user mentions FULLY (no limit/offset) before decomposing — you need full context before you split the work.
2. Decompose
Break the question into composable research areas. ultrathink about the patterns, connections, and architecture the user is really after.
3. Spawn parallel sub-agents
Run concurrently — each a documentarian (tell them WHAT to find, not how; remind them to describe, not evaluate):
codebase-locator— WHERE things livecodebase-analyzer— HOW the code workscodebase-pattern-finder— existing patterns to point at
Use Explore for a question too diffuse for targeted locators; WebSearch only if the user explicitly asks.
4. Synthesize
Wait for ALL agents before proceeding. Connect findings across components with file:line evidence, answering the user's actual question. Stay on synthesis — don't deep-read files in main context.
A sub-agent's finding is a claim, not evidence. Before a claim enters the doc, open its file:line and confirm the code says what the agent reported. A same-model sub-agent buys context separation, not independent verification.
5. Write the doc
Write to specs/research/YYYY-MM-DD-<kebab-slug>.md (create the dir) using the structure and metadata commands in `RESEARCH-TEMPLATE.md`. Fill the metadata from real command output first — never write placeholder values.
6. Permalinks, then present
On a pushed branch or main, swap local refs for GitHub permalinks (gh repo view --json owner,name). Then give the user the doc path and a concise summary with key references. Append follow-ups to the same doc under a ## Follow-up Research <date> heading and bump last_updated.
Done when the doc answers the question purely descriptively (zero recommendations), every claim carries a file:line, and the metadata is real.
Security
- Security scanPassed, 90 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/research/SKILL.md
Popularity
- 7-day trend score
- 0
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/2btwist/dotagents/research)