incorporating-research

Incorporating Research

Guidelines: See CLAUDE.md for writing standards. Last Updated: 2026-01-26

Purpose

This skill handles incorporating external research (Perplexity, web searches, ChatGPT outputs, etc.) into existing documents without breaking voice, structure, or adding irrelevant content.

Core principle: Triage first, confirm scope, then edit. Strip artifacts, preserve voice.

When to Use

Invoke when user says:

• "Incorporate this feedback"
• "Merge this into the doc"
• "Add this Perplexity/research output"
• "Update the doc with this"
• User pastes a block of external content after discussing a document

Red flag: User says "incorporate" but might mean "review for quality" — clarify if ambiguous.

The Workflow

Step 1: Identify Target Document

Ask if unclear:

• "Which file should I update with this research?"
• If multiple candidates exist, list them and ask

Confirm:

• "I'll incorporate this into [path/to/file.md]. Correct?"

Step 2: Triage the Input

Separate signal from noise:

Category	What to Look For	Action
New content	Relevant facts, data, examples	Extract for incorporation
Irrelevant content	Wrong topic, unrelated sections	Strip completely
Citation artifacts	[1][2][3], "Sources:", URLs	Strip unless explicitly requested
Formatting quirks	Perplexity/ChatGPT styling	Normalize to doc style
Hallucinations	Suspicious claims, outdated info	Flag for user verification

Output triage summary:

Triage Results:
├── New content: [Brief description]
├── Irrelevant: [What's being stripped]
└── Artifacts: [Citation numbers, source sections, etc.]

Step 3: Map to Existing Structure

Read target document first. Determine: new section, inline addition, appendix, or replacement. Flag conflicts/duplicates.

Step 4: Confirm Scope Before Editing

NEVER edit without confirmation. Show: changes to [file], what to strip, concerns. Wait for approval.

Step 5: Preserve Voice and Strip Artifacts

Match: header levels, list style, tone, code/link formatting. Strip: citation numbers [1], source/reference sections, AI metadata ("Based on X sources", "Here's a summary"), unrelated content.

Step 6: Post-Incorporation Summary

Report: what was added and where, what was stripped, any concerns.

Red Flags to Catch

Red Flag	Example	Action
Topic mismatch	Kamailio versions in hardware procurement doc	Strip completely, ask user
Ambiguous intent	"Incorporate this" (could mean review)	Clarify: "Add to doc or review for quality?"
Hallucinations	Outdated versions, wrong dates	Flag: "This claims X, verify?"
Contradictions	New content conflicts with existing	Flag before incorporating
Duplicate content	Research repeats what's already there	Ask: "This duplicates section Y. Replace or skip?"

Common Mistakes to Avoid

❌ Reviewing instead of incorporating — User says "incorporate" but wants quality review ❌ Breaking document voice — Pasting ChatGPT-style prose into technical doc ❌ Leaving citation artifacts — [1][2][3] scattered throughout ❌ Adding irrelevant content — Including unrelated sections from paste ❌ Editing without confirmation — Making changes before user approves scope

✅ Triage → Map → Confirm → Edit → Summarize

Example Interaction

User: "Incorporate this Perplexity output into the hardware doc"

Agent:

I'll incorporate this into `docs/hardware-procurement.md`. Let me triage the input first.

Triage Results:
├── New content: GPU specs for H100 vs A100
├── Irrelevant: Section on Kamailio SIP server versions (wrong topic)
└── Artifacts: Citation numbers [1][2][3], "Sources" section with 5 URLs

I'll make these changes:

1. Add GPU comparison table to "Hardware Options" section
2. Strip Kamailio section (unrelated to hardware)
3. Strip citation numbers and Sources section

Concerns:
- Claims H100 has 80GB VRAM, but some models have 96GB. Verify?

Proceed?

User: "Yes, use 80GB for now"

Agent: [Makes edits]

Incorporated into docs/hardware-procurement.md:

Added:
- GPU comparison table (H100 vs A100) in "Hardware Options" section

Stripped:
- Kamailio SIP server section (unrelated)
- Citation numbers [1]-[5]
- Sources section with URLs

Concerns:
- None (H100 VRAM confirmed as 80GB for standard model)

Companion Skills

• eliminating-ai-slop: Clean up pasted text · detecting-ai-slop: Score research quality
• expert-interviewer: Structured knowledge extraction · perplexity-research: Web research

Resource	Purpose
CLAUDE.md	Writing standards (no AI slop)
eliminating-ai-slop	Use after incorporating to clean up pasted text
detecting-ai-slop	Score research quality before incorporating

grep -n "TODO.*verify" draft.md  # find unverified claims before publish

Failure Modes

Failure	Fix
Copy-paste without attribution	Cite the source inline
Stale sources	Check publication date before incorporating
AI-slop injection	Run eliminating-ai-slop after pasting