name: blog-research description: Research legal technology topics for blog posts on alt-counsel (Ang Hou Fu's blog). Use when the user asks to research topics, find sources, fact-check claims, gather statistics, or find expert opinions for blog content. Prioritizes Singapore/ASEAN perspectives and flags US/EU-centric information. Outputs research findings to research.md in the post folder with proper citations.

Blog Research Skill

Research legal technology topics for alt-counsel blog posts with Singapore/ASEAN prioritization and jurisdictional awareness.

When to Use This Skill

Trigger this skill when the user requests:

"Research [topic] for my blog post"
"Find sources on [legal tech topic]"
"Fact-check: [claim]"
"Get statistics on [topic]"
"Find expert opinions on [subject]"
Any research task related to blog content creation

Research Workflow

Step 0: Verify Research Readiness

Before beginning research, confirm the post has:

Before anything else — check the blog archive: Use the ghost_search MCP tool (full-text search across posts, pages, tags) to check if any existing posts on this topic already exist. For filtered queries, use ghost_post_list with NQL filters. This prevents: assuming a post exists when it doesn't, repeating content already covered, and missing series context. Do NOT assume prior posts exist based on topic relevance alone.

Required inputs:

[ ] Locked thesis - One-sentence argument the post will make
[ ] Section outline - 3-5 main sections with their specific claims
[ ] Evidence gaps - Specific claims that need external validation

Stop and ask user if missing:

"I see you want research on [topic], but I don't have your post outline yet."
"Which specific claims in your post need evidence?"
"What are you trying to prove with this research?"

Why this matters: Research without structure leads to gathering interesting information that doesn't make the final cut. Lock the argument first, research second.

Step 1: Understand the Research Need

Critical question: What needs external validation vs what's personal experience?

Ask user to categorize each post section:

Personal experience (✍️) - No research needed, this is your story
Needs validation (🔍) - Your claim needs external support
Needs context (🌍) - Your experience needs broader context

Example:

Section 2: "My 3-page prompt experience" = ✍️ (don't research this)
Section 3: "Agents changed in Sept 2025" = 🔍 (needs validation)
Section 4: "Singapore firms struggling too" = 🌍 (needs regional context)

Only research the 🔍 and 🌍 sections.

If the post folder is ambiguous, ask the user to specify (e.g., posts/contract-automation-tools/).

Step 2: Read Reference Files

Before conducting research, read:

references/trusted-sources.md - Know which sources to prioritize
references/regional-considerations.md - Understand Singapore/ASEAN vs US/EU differences

These files provide critical context for effective research.

Step 2.5: Generate Research Brief for Approval

Before conducting searches, output a research brief:

# Research Brief: [Post Title]

**Locked thesis:** [One sentence]

**Evidence needed:**
1. Section X: [Specific claim] 
   - Searching for: [What evidence would support this]
   - Stop criteria: [When I have enough]
   
2. Section Y: [Specific claim]
   - Searching for: [Evidence needed]
   - Stop criteria: [When sufficient]

**Not researching:**
- Section Z: Personal experience (no validation needed)

**Estimated searches:** [Number based on sections]
**Stop when:** All sections adequately supported OR regional sources exhausted

Ask the user only if the brief changes the thesis or the budget. Otherwise post the brief and start searching — it is a plan, not a permission request. Report what you found against it.

Step 3: Conduct Research

Use web_search following this priority order:

Search Strategy:

Singapore/ASEAN-first search - Add "Singapore" or "ASEAN" qualifiers to queries
Evaluate regional results - Check if findings are regionally relevant
Expand to global sources - If regional sources insufficient, search without geographic qualifiers
Flag jurisdictional concerns - Note when findings are US/EU-centric

Search Tactics:

Start with 3-5 searches minimum (regional → global)
Use web_fetch to read full articles from promising results
Extract specific data points, quotes, and citations
Cross-reference information across multiple sources

Search depth — go wide. There is no useful cap on searches; a thin evidence base costs a rewrite, an extra search costs a minute. Per section claim, stop when it is supported by 2+ credible sources or regional sources are exhausted — but follow leads that look promising even when they were not in the brief, and record them.

The cap is on what crosses into the pitch, not on what you find. research.md is unbounded. Per drafting session, at most 3 findings may change the pitch's beats — pick the three with the highest claim-per-word and log the rest in research.md under "Available, not used." A finding that adds a thread without replacing one is not a finding, it is scope. (History: one post accumulated ~12 threads in a session because everything research found entered the pitch.)

Red flag: If the outline still has evidence gaps after wide searching, the problem is likely the outline (claims too broad/unsupported), not insufficient research.

Step 3.5: Read the artifact, not the log

When a source is a discussion, chat log, or session transcript, it is evidence of what someone said, not of what they decided. If a message produced an artifact — a design doc, an ADR, a commit, a merged PR — open the artifact and cite that. Pitch v6 of legal-oss-maintainer failed on exactly this: research recorded four soft bullets from a chat message and missed the design principle entirely, because the principle was in the design doc the message produced, four commands away in the repo, never opened. It cost a full draft. Interrogating your framing cannot tell you that you read the wrong file.

Step 4: Evaluate Sources

Apply source evaluation criteria from references/trusted-sources.md:

Recency - Prefer sources from last 6-24 months for trends/pricing
Regional relevance - Singapore/ASEAN > Global > US/EU-specific
Credibility - Established publications > vendor blogs
Specificity - Actual data/examples > general claims
Independence - Journalism > vendor marketing

Step 5: Apply Regional Lens

For EVERY finding, assess:

Is this Singapore/ASEAN-relevant?
Is this US/EU-centric? If so, what differs locally?
What's the alt-counsel angle (resource-constrained perspective)?

Flag jurisdictional concerns using the protocol from references/regional-considerations.md.

Step 6: Map Findings to Post Structure

For EACH section in the user's outline, identify:

Section X claim: [The specific point this section makes]
Evidence needed: [What this claim requires for support]
Findings that support: [Which research findings apply here]
Findings that don't fit: [Interesting but not relevant to this section]

Red flag if: You have 5+ findings that don't map to any section. Action: Ask user: "I found [X], but it doesn't support your outlined sections. Should I continue researching or is this exploratory?"

Quality check: Every finding should answer "Which section needs this?"

Step 7: Format Research Output

Load assets/research-template.md and populate it with findings:

Required sections:

Summary - 2-3 sentence executive summary
Key Findings - Detailed findings with sources and regional context icons
Statistics & Data Points - Specific data with citations
Expert Quotes - Direct quotes from credible sources
Jurisdictional Flags - US/EU-centric content flagged with explanations
Alt-Counsel Angle - How findings relate to blog's resource-conscious mission
Additional Sources - Links for follow-up research

Citation format:

Source: Article Title - [Publication]
Date: [Publication date]
Regional Context: ✅ Singapore/ASEAN | ⚠️ US/EU-centric | 🌍 Global

Step 8: Save Research Output

Create research.md in the specified post folder:

Path format: posts/[post-folder-name]/research.md
Always confirm the post folder path with the user if ambiguous

Regional Prioritization Rules

Always prioritize:

Straits Times, Channel News Asia for Singapore news
Artificial Lawyer for global legal tech (but flag when US/EU-centric)
Singapore government sources (Courts, Legal Service, IMDA)
ASEAN business and legal publications

Always flag when:

Information is US/EU-specific (ethics rules, regulations, market dynamics)
Pricing assumes enterprise budgets
Solutions require infrastructure uncommon in ASEAN
No Singapore/ASEAN-specific information available

Quality Standards

Good research includes:

Minimum 5-8 quality sources cited
Mix of regional and global perspectives
Specific statistics with dates
Expert quotes from credible sources
Clear jurisdictional flags where relevant
Alt-counsel angle articulated

Bad research avoids:

Vendor marketing without independent validation
Outdated sources (>2 years old for trends)
Pure US content without regional context
Vague claims without specific examples
Missing citations or broken links
Unverified statistics from memory (always web-search to confirm data before writing it into drafts)

Example Research Queries

Fact-check query:

"Fact-check: Contract lifecycle management systems typically cost $50K+ for enterprises"

Workflow: Search for CLM pricing, gather data points from multiple sources, note regional pricing if available, flag if US-centric pricing, provide alt-counsel perspective on accessible alternatives.

Topic research query:

"Research AI document review tools for small legal teams"

Workflow: Search for AI document review tools, prioritize solutions for small teams, compare pricing tiers, find Singapore/ASEAN adoption examples, flag enterprise-only solutions, identify practical alternatives.

Statistics query:

"Get statistics on legal tech adoption in Singapore law firms"

Workflow: Search Singapore-specific legal tech adoption data, check Law Society publications, look for regional surveys, note data gaps, compare to global trends where relevant.

Recherche pour blog legaltech

Recommandé pour