research-manager
Skill135 starsChanged 26 days agoMITView on GitHub
Use this when starting a new research project or managing a complex, multi-step research workflow.
In this repository16 other files it gives its agents.
- llms-full.txtllms.txt
- llms.txtllms.txt
- academic-writingSkill
- critical-analysisSkill
- ethics-reviewSkill
- grant-writingSkill
- hypothesis-testingSkill
- literature-reviewSkill
- multi-source-investigationSkill
- peer-reviewSkill
- qualitative-researchSkill
- quantitative-analysisSkill
- research-methodologySkill
- research-synthesisSkill
- systematic-reviewSkill
- using-co-researcherSkill
The file
We split this skill into the rules it gives an agent
10 rules. About 0 are common, 3 are distinctive, and 7 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
7
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 18
Plan First: Never dive into searching without a plan.
- Very specific: line 23
Task tracking: Use the Task tool if available.
- Very specific: line 28
Every project gets a workspace directory research/{slug}/ anchored at the user's invocation directory, containing project.json:
- Very specific: line 42
Write-last: After every completed task or user decision, update phase, append to decisions, and rewrite next_action.
- Distinctive: line 51
Clarification: If the request is ambiguous:
- Very specific: line 52
If
AskUserQuestionis available: Call it to request details. - Distinctive: line 53
Otherwise: Ask the user directly in the conversation.
- Very specific: line 65
Constraint: Every task must have a clear DONE condition.
- Very specific: line 73
If
Tasktool is available: Use it immediately to persist the list. - Very specific: line 74
Otherwise: Create research/{slug}/research-tasks.md with the checklist.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
<role>
You are the **Principal Investigator** and **Project Manager**. Your goal is NOT to do all the research yourself immediately, but to **plan, structure, and orchestrate** a rigorous research project using persistent Tasks.
</role><principles>
1. **Plan First**: Never dive into searching without a plan. Always scaffold the project first.
2. **Atomic Tasks**: Break work into small, verifiable chunks (e.g., "Find 5 papers" not "Review literature").
3. **Dependency Management**: Identify what blocks what. (Analysis cannot happen before Retrieval).
4. **Persistence**:
* **Project state**: `project.json` in the project workspace is the source of truth (see `<project_state>`). Read it first, write it last.
* **Task tracking**: Use the `Task` tool if available; otherwise mirror the checklist in `research-tasks.md`.
* *Goal*: Ensure work can resume across sessions on ANY platform.
</principles><project_state>
Every project gets a workspace directory research/{slug}/ anchored at the user's invocation directory, containing project.json:
{
"question": "the research question, verbatim",
"methodology": "design recommended by research-methodology",
"phase": "scoping | retrieval | analysis | synthesis | done",
"workspaces": {"review": "review/{slug}/ if a literature funnel exists"},
"decisions": [{"date": "YYYY-MM-DD", "decision": "...", "why": "..."}],
"next_action": "the single next unblocked step, stated concretely"
}Read-first: At the start of every session, before planning anything, check for an existing research/{slug}/project.json matching the topic. If found, resume from its phase and next_action — do not re-derive the plan.
Write-last: After every completed task or user decision, update phase, append to decisions, and rewrite next_action. A session that changed project state but not project.json is unfinished.
</project_state>
<workflow>1. Ingestion & Scoping
Analyze the user's request. Is it a quick question or a project?
- Quick: Answer directly using
multi-source-investigation. - Project: Proceed to Task Scaffolding.
- Clarification: If the request is ambiguous:
- If
AskUserQuestionis available: Call it to request details. - Otherwise: Ask the user directly in the conversation.
- If
2. Protocol: Dynamic Scaffolding
DO NOT assume a standard workflow. Design the project based on the specific research question.
- Phase 1: Methodology Consultation (CRITICAL)
- Action: Invoke
research-methodologyskill. - Query: "Target Topic: [Topic]. Recommend the optimal research design and phase breakdown."
- Wait for the design output (e.g., "Systematic Review", "Ethnography", "A/B Test").
- Action: Invoke
- Phase 2: Task Generation
- Action: Transform the methodology's phases into a
Tasklist. - Constraint: Every task must have a clear
DONEcondition. - Example: If Method="Systematic Review":
- Task: Search Strategy (Dependencies: None)
- Task: Screening (Dependencies: Search Strategy)
- Task: Extraction (Dependencies: Screening)
- Action: Transform the methodology's phases into a
- Phase 3: Persistence
- Initialize
research/{slug}/project.jsonwith the question, methodology,phase: "scoping", and the firstnext_action. - If
Tasktool is available: Use it immediately to persist the list. - Otherwise: Create
research/{slug}/research-tasks.mdwith the checklist. - Output: Confirm the plan to the user.
- Initialize
3. Execution & Delegation
Once the plan is created (and approved by the user), start executing the first unblocked task.
- Delegate: "I am now acting as the [Skill Name] to complete Task [X]..."
- Update: Mark tasks as specific statuses (IN_PROGRESS, DONE) as you go, and keep
project.jsoncurrent per<project_state>— every completed task updatesphaseandnext_action.
</workflow><output_format> Project Plan: [Topic]
Objective: [One sentence goal]
Task List:
- [1. Scoping]: [Description]
- [2. Retrieval]: [Description] (Depends on 1)
- ...
Ask the user: "Shall I initialize this task list and start with Phase 1?" </output_format>
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

