---

name: "cirf" version: "1.0.0" description: "Interactive crypto deep-research framework with agent-agent and human-agent collaboration for superior research outcomes" author: name: "Kudō" social: "https://x.com/kudodefi" homepage: "https://github.com/kudodefi/cirf" metadata: emoji: "🔬" category: "research"

CIRF - Crypto Interactive Research Framework

FRAMEWORK OVERVIEW

CIRF is a systematic crypto research framework with:

• 4 Specialized Agents with distinct expertise
• 17 Research Workflows covering market, project, and technical analysis
• Workspace System for project isolation and output management

Framework Location: ./framework/ Workspaces Location: ./workspaces/

---

AGENT REGISTRY

🔬 Research Analyst

• File: ./framework/agents/research-analyst.yaml
• Expertise: Market intelligence, project fundamentals, investment synthesis
• Workflows: All research workflows (sector, project, competitive, trend, traction, tokenomics, etc.)
• Use for: Market analysis, project evaluation, investment thesis

⚙️ Technology Analyst

• File: ./framework/agents/technology-analyst.yaml
• Expertise: Architecture assessment, security analysis, technical evaluation
• Workflows: technology-analysis
• Use for: Smart contract review, protocol architecture, technical due diligence

✍️ Content Creator

• File: ./framework/agents/content-creator.yaml
• Expertise: Research-to-content transformation, multi-platform optimization
• Workflows: create-content
• Use for: Converting research into threads, articles, reports

✓ QA Specialist

• File: ./framework/agents/qa-specialist.yaml
• Expertise: Quality validation, critical review, bias detection
• Workflows: qa-review
• Use for: Reviewing outputs, challenging assumptions, finding gaps

---

WORKFLOW REGISTRY

Setup & Planning

ID	Description
framework-init	First-time user configuration
create-research-brief	Define research scope & objectives

Market Intelligence

ID	Description
sector-overview	Sector structure, dynamics, key players
sector-landscape	Ecosystem mapping, player categorization
competitive-analysis	Head-to-head project comparison
trend-analysis	Trend identification & forecasting

Project Fundamentals

ID	Description
project-snapshot	Quick project overview
product-analysis	Product mechanics, PMF, innovation
team-and-investor-analysis	Team background, investor quality
tokenomics-analysis	Token economics, sustainability
traction-metrics	Growth, retention, unit economics
social-sentiment	Community health, sentiment

Technical & Quality

ID	Description
technology-analysis	Architecture, security, code quality
qa-review	Validation, bias detection, gap analysis

Content & Flexible

ID	Description
create-content	Multi-format content package
open-research	Custom research on any topic
brainstorm	Ideation and exploration

---

ORCHESTRATION PROTOCOL

STEP 1: UNDERSTAND

Parse user's research goal:

• Subject: Project, sector, or topic
• Outcome: Investment decision, market understanding, content, etc.
• Scope: Depth, time horizon, specific focus areas

If unclear, ask clarifying questions before proceeding.

STEP 2: PLAN & BUILD TASK QUEUE

Create task queue with this structure:

TASK QUEUE: {research_goal}
Workspace: {workspace_id}
Created: {timestamp}

| Task ID | Workflow | Agent | Status | Depends On |
|---------|----------|-------|--------|------------|
| T1 | {workflow} | {agent} | pending | - |
| T2 | {workflow} | {agent} | pending | - |
| T3 | {workflow} | {agent} | pending | T1, T2 |
| ... | ... | ... | ... | ... |

Status values: pending → in_progress → completed | failed

Dependency rules:

• Tasks with no dependencies can run in parallel
• Tasks wait for all dependencies to complete
• Group tasks by agent for efficient spawning

STEP 3: CONFIRM

Present plan to user:

RESEARCH PLAN: {goal}

Will execute {n} tasks across {n} agents:

PHASE 1 (Parallel):
• T1: {workflow} ({agent})
• T2: {workflow} ({agent})

PHASE 2 (After Phase 1):
• T3-T5: {description}

PHASE 3 (Final):
• T6: {workflow}

Proceed? [yes / adjust / questions]

Wait for user approval before execution.

STEP 4: EXECUTE WITH PROGRESS TRACKING

4.1 Spawn agents by grouping:

• Group all tasks assigned to same agent
• Spawn one sub-agent per agent-type with their full task list
• Sub-agent executes tasks sequentially, respecting dependencies

4.2 Update progress after each task completion:

PROGRESS UPDATE
═══════════════════════════════════════════

Overall: ████████░░░░░░░░ 50% (4/8 tasks)

✅ COMPLETED:
• T1: {workflow} → {key finding summary}
• T2: {workflow} → {key finding summary}

🔄 IN PROGRESS:
• T3: {workflow} ({agent})

⏳ PENDING:
• T4: {workflow} (waiting for T3)
• T5: {workflow} (waiting for T3, T4)

═══════════════════════════════════════════

STEP 5: CONSOLIDATE

When all tasks complete:

1. Collect outputs from ./workspaces/{workspace}/outputs/
2. Synthesize key findings across all analyses
3. Generate final deliverable
4. Present to user with confidence level and recommendations

---

SUB-AGENT SPAWN PROTOCOL

Spawn agent with assigned task queue:

SPAWN: {agent-type}

IDENTITY:
Read and embody: ./framework/agents/{agent-id}.yaml

WORKSPACE:
./workspaces/{workspace-id}/

TASK QUEUE:
| Task ID | Workflow | Status | Depends On |
|---------|----------|--------|------------|
| T1 | {workflow-1} | pending | - |
| T3 | {workflow-2} | pending | T1 |
| T6 | {workflow-3} | pending | T3 |

EXECUTION PROTOCOL:
For each task in queue (in dependency order):

1. CHECK DEPENDENCIES
   - If dependencies not completed, wait or skip to next available task

2. UPDATE STATUS
   - Mark task: in_progress

3. READ WORKFLOW FILES
   - ./framework/workflows/{workflow-id}/workflow.yaml
   - ./framework/workflows/{workflow-id}/objectives.md
   - ./framework/workflows/{workflow-id}/template.md

4. READ CONTEXT
   - ./framework/core-config.yaml (user preferences)
   - ./workspaces/{workspace}/workspace.yaml (project context)

5. EXECUTE
   - Follow methodology in objectives.md
   - Apply output structure from template.md

6. SAVE OUTPUT
   - Path: ./workspaces/{workspace}/outputs/{workflow-id}/{workflow-id}-{date}.md

7. UPDATE STATUS
   - Mark task: completed

8. REPORT TO ORCHESTRATOR
   - task_id
   - output_path
   - key_findings (3-5 bullets)
   - confidence (high/medium/low)

COMPLETION:
When all assigned tasks done, report final summary to orchestrator.

---

WORKSPACE MANAGEMENT

Initialize Workspace

Before execution, ensure workspace exists:

1. Check if ./workspaces/{workspace-id}/ exists
2. If not, create workspace:
   • Copy template from ./framework/_workspace.yaml
   • Create documents/ directory
   • Create outputs/ directory
3. Update workspace.yaml with project metadata

Output Structure

workspaces/{project}/
├── workspace.yaml          # Project configuration
├── documents/              # Source materials, references
└── outputs/
    ├── sector-overview/
    │   └── sector-overview-{date}.md
    ├── project-snapshot/
    │   └── project-snapshot-{date}.md
    ├── tokenomics-analysis/
    │   └── tokenomics-analysis-{date}.md
    └── ...

---

CONFIGURATION

Read ./framework/core-config.yaml at initialization:

Field	Purpose
status.initialized	If false, run framework-init first
user.name	For personalized communication
language.communication	Interaction language (en/vi/zh)
language.output	Output document language (en/vi/zh)
user.currency	Currency format preference
user.date_format	Date format preference

---

ERROR HANDLING

If a task fails:

1. Mark status: failed
2. Log error: Record reason in task queue
3. Notify orchestrator with error details
4. Determine action:
   • Retry: Attempt task again (max 2 retries)
   • Skip: Continue if non-critical (mark skipped)
   • Pause: Stop execution and ask user for guidance

⚠️ TASK FAILED
═══════════════════════════════════════════

Task: T3 - tokenomics-analysis
Agent: Research Analyst
Error: {error description}

Options:
1. Retry task
2. Skip and continue
3. Pause and review

═══════════════════════════════════════════

---

QUALITY STANDARDS

All research outputs must:

• Source citations: Include source name, date, and credibility level
• Information classification: Distinguish verified facts vs credible reports vs speculation
• Confidence levels: State confidence (high/medium/low) for conclusions
• Language compliance: Follow language.output setting
• Terminology: Preserve English crypto domain terms regardless of output language
• Structure: Follow workflow template.md format

---

QUICK REFERENCE

Activation Checklist

1. Read this SKILL.md
2. Check ./framework/core-config.yaml → status.initialized
3. If not initialized, run framework-init workflow
4. Ready to accept research goals

Research Execution Flow

User Goal → Understand → Plan → Confirm → Execute → Consolidate → Deliver

Key Paths

Resource	Path
Agents	./framework/agents/{agent-id}.yaml
Workflows	./framework/workflows/{workflow-id}/
Config	./framework/core-config.yaml
Workspace Template	./framework/_workspace.yaml
Workspaces	./workspaces/{workspace-id}/
Outputs	./workspaces/{workspace-id}/outputs/{workflow-id}/