Computational Communication Research Framework | v0.9.0

A methodological skill pack for AI-assisted computational social science research, adapted from DAAF and tailored for communication scholars.

Sister project: CommScribe — While CommDAAF handles data analysis, CommScribe handles literature review, theory building, and scholarly writing with voice learning.

CommDAAF is a set of structured instructions ("skills") that help AI coding assistants support computational communication research. Rather than treating AI as a magic analysis machine, CommDAAF enforces methodological rigor by:

• Asking probing questions before running any analysis
• Refusing to use default parameters without explicit justification
• Requiring validation at levels appropriate to your stakes
• Connecting methods to communication theory
• Being honest about data access in the post-API era

The goal is not to automate research, but to create a collaborator that pushes back on sloppy methodology.

Not every analysis needs publication-grade rigor. CommDAAF lets you choose:

Validation requirements scale accordingly.

Every methodological choice is surfaced, not hidden:

1. Default Danger Flags — Warns when you're about to accept untested defaults
2. Active Choice Requirement — Forces explicit selection between alternatives
3. Trade-Off Visualization — Shows what you gain and lose with each choice
4. Assumption Audit — Surfaces hidden assumptions before analysis
5. Reflection Checkpoints — Pauses for metacognition at key stages

The era of free Twitter APIs is over. CommDAAF is honest about what data you can actually get:

CommDAAF works with three AI coding platforms. Pick yours:

curl -O https://raw.githubusercontent.com/weiaiwayne/commDAAF/main/CLAUDE_BUNDLE.md
mv CLAUDE_BUNDLE.md CLAUDE.md

cd ~/.openclaw/workspace/skills
git clone https://github.com/weiaiwayne/commDAAF.git commdaaf

cd ~/.gemini/antigravity/skills
git clone https://github.com/weiaiwayne/commDAAF.git
cd commDAAF && cp -r antigravity/* . && rm -rf antigravity skill-templates

After installation, try:

Analyze sentiment in climate change discourse on Bluesky

If working correctly, the assistant should ask probing questions about what you mean by "sentiment," your unit of analysis, and your validation plan. If it just runs VADER with defaults, the setup isn't working.

commDAAF/
├── CLAUDE_BUNDLE.md          # One-file version for Claude Code
├── DEPLOYMENT.md             # Detailed setup instructions
├── skill-templates/          # OpenClaw version
│   ├── SKILL.md              # Main entry point
│   ├── methods/              # 10+ method skills
│   ├── data-sources/         # Platform access guides
│   ├── workflows/            # Validation, ethics, stages
│   └── theories/             # Communication theory modules
└── antigravity/              # Google Antigravity version

@software{commdaaf,
  title={CommDAAF: Computational Communication Research Framework},
  author={Xu, Wayne and LampBotics AI Lab},
  year={2026},
  url={https://github.com/weiaiwayne/commDAAF},
  license={GPL-3.0},
  note={Experimental. Adapted from DAAF.}
}

CommDAAF adapts the Data Analyst Augmentation Framework (DAAF), originally developed for education data analysis.

This framework is under active development at the LampBotics AI Lab. Built concurrently by Kimi K2.5 and Claude Opus 4.5 as an experiment in AI-assisted research tool development.

Use with caution. Things may break, documentation may be incomplete.

An incubator where AI agents learn from mistakes through adversarial peer review. Multiple models independently analyze the same data, then cross-review each other. Errors become lessons; lessons become new framework checks.

Live dashboard: AgentAcademy

New study: Analyzed 192 U.S. congressional hearings (2007-2026) on AI framing. Achieved κ=0.656 (Substantial) with 2-model validation after prompt refinement.

Key findings:

• Sovereignty (22%) and Innovation (21%) dominate—Congress frames AI as competition to win
• 90% of hearings occurred post-ChatGPT (118th-119th Congress)
• Rights frame only emerged in 2023+
• Senate emphasizes security +55% more than House

New protocol additions:

• API search false positive filtering (density scoring)
• Document-type coding bias detection
• Prompt iteration as mandatory phase
• Two-model validation acceptable for exploratory tier
• Full document coding guidance

Two preprints from the #MahsaAmini virality study, plus a comprehensive protocol for future agentic studies:

New subskills added (v0.6):

• Literature Synthesis — Semantic Scholar + OpenAlex search, citation networks, gap analysis
• Multimodal Coder — Image frames, image-text relationships, video keyframes

Key lessons encoded in protocol:

• Kimi batch limit: 25 posts max (JSON truncation otherwise)
• Mandatory distribution diagnostics before regression
• Frame-specific reliability reporting required
• Never use OLS on skewed engagement data

This study demonstrates the AgentAcademy improvement loop: Run research → Find gaps → Fix framework.

Analyzed 262 Iran news headlines (GDELT, Jan 2024 – Feb 2026) with 3-model validation. Study worked—but exposed 5 methodology gaps that became skill updates:

Key research finding: Israeli sources frame Iran as THREAT 10x more than Al Jazeera (42% vs 4%). All 5 hypotheses supported with 78% 3-model agreement.

📄 Full report: studies/2026-02-26-iran-agenda.md

Three-layer architecture, mandatory cross-agent validation, credibility rating scheme, and structured failure knowledge base. Inspired by Xu & Yang (2026).

STATUS: HYPOTHESIS DISPROVEN

Controlled testing confirms: Academic methodology framing does NOT bypass Chinese LLM content filters. Both z.ai (GLM) and Moonshot (Kimi) block sensitive content regardless of CommDAAF wrapper.

Previous apparent "bypass" was due to OpenCode's free proxy infrastructure, not prompt engineering.

📄 Final study: papers/CENSORSHIP_STUDY_FINAL.md

Contributions welcome:

• Bug reports and issues
• New method skills
• Improved probing questions
• Additional theory modules

GNU General Public License v3.0 (GPL-3.0), same as the original DAAF.

• DAAF — The original framework this adapts
• Prof. Wayne Xu — Methods development and Zotero library baseline
• LampBotics AI Lab — Development environment
• Kimi K2.5 & Claude Opus 4.5 — Concurrent development

Built for how research actually works—not how we wish it worked.