Briefing Skill — Understanding Workbench

What It Does

The Briefing skill bridges the gap between analytical data and human decisions. It takes what Ingestion discovered about a corpus and translates it into a conversation that helps the human information architect decide what to do next.

This is the most distinctly human-facing skill in the pipeline. Its output is a briefing document — not a recommendation, but a structured set of options with honest tradeoffs.

Three-Phase Workflow

Phase 1

Reference Interview

1-2 clarifying questions. Work backward from the question to the actual need.

Phase 2

Capabilities Briefing

Translate statistics into implications. Show what's there and what's possible.

Phase 3

Collaborative Triage

2-3 options with effort, output, and tradeoffs. Suggest, don't decide.

Phase 1: The Reference Interview

A reference librarian knows that the first question asked is rarely the actual need. People state tasks, but tasks emerge from needs, and needs emerge from situations. The reference interview works backward:

Situation → Need → Task → Question Asked

Ask at most 1-2 questions. Listen for signals about what they're really trying to accomplish: navigation, migration, governance, compliance, search improvement, content audit, or research.

If they want to skip ahead, let them. This is not an interrogation — it's a professional courtesy.

Phase 2: The Capabilities Briefing

Present what the Collection Record tells you, translated into plain language. This is not a statistics dump — it's a capabilities briefing that respects the person's intelligence.

Structure: corpus at a glance, what's already organized, what analysis can reveal (as a table of options), quality notes, and what additional information would unlock more options.

The key discipline: translate statistics into implications. "Moderate vocabulary diversity" becomes "typical for focused domain content — will cluster well." Numbers serve the narrative, not the other way around.

Phase 3: Collaborative Triage

Present 2-3 concrete options. Each option includes what you'd do, the effort level, what you'd get, and what you'd give up. End with a suggestion and rationale, framed as a recommendation — not a decision.

"For a corpus this size, Option B gives you the best balance
of accuracy and effort. But if you're planning a full redesign,
Option C is worth the extra time. Your call."

Anti-Patterns

✕

Statistics dump

Don't present raw numbers without translation. Every statistic should be accompanied by its implication.

✕

Single recommendation

Always present options. People want to understand their choices, not be told what to do.

✕

Decision paralysis

2-3 options maximum. More than that overwhelms rather than empowers.

✕

Jargon

Explain in terms of what they'll get, not in terms of the method used to get it.

✕

Gatekeeping

If they want to skip the interview and just run analysis, support that. Respect their judgment.

How to Use

Via Claude:

"I have a collection record for [corpus]. Generate a briefing
with options for what we should do next."

Via command line:

python skills/librarian-briefing/scripts/generate_briefing.py \
  --collection-record collection-record.json \
  --purpose "Content audit for website redesign" \
  --audience "Content strategists and web team" \
  --output briefing.md

After Triage — Dispatch Table

If They Choose	Next Skills to Run
Topic-First	topic_modeling → vocabulary_analysis
Classification-First	librarian-enrichment → manual review
Full Faceted	topic_modeling → vocabulary_analysis → taxonomy_builder → concept_graph
"Just show me what's there"	librarian-similarity → Organizer Console

Corpus Briefing