agentleFS
samsational-codesamsational-code/business-skills/presentation-transcript-extractor/SKILL.md

presentation-transcript-extractor

Skill0 starsChanged 2 months agoMITView on GitHub

Turn a long formal presentation-plus-Q&A transcript — earnings call, all-hands or town hall, investor day, product keynote, press conference, webinar, conference session, or board presentation — into a structured memo: key metrics with period-over-period changes exactly as stated, what leadership emphasized, every audience question with an answer-quality rating, where speakers were direct vs. evasive, and a condensed transcript at roughly half the original length. For analysts, investors, employees, journalists, competitive-intelligence teams, customers, partners, and board members who need to parse what was actually said. Trigger on "process this transcript," "summarize the earnings call," "what did leadership say at the all-hands," "pull the Q&A," "extract the metrics from this keynote," "memo on the investor day," "what did they commit to," "where were they evasive," or whenever the user provides a long transcript-like text with speaker labels, audience questions, and metrics. Do NOT use for ordinary meetings or sales calls that produce action items — use the call-transcript-followup skill. Do NOT use for formal board or committee minutes — use the meeting-minutes-generator skill. Do NOT use for comparing two versions of a statement or document — use the document-redline-diff skill.

The file

We split this skill into the rules it gives an agent

11 rules. About 0 are common, 10 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈10

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If the transcript is very long (common — these sessions often run 10,000+ words), read it in full before beginning analysis.

    line 48
  2. Distinctive:

    Before starting the memo, identify and record at the top:

    line 52
  3. Distinctive:

    Identifier: if applicable (ticker symbol for a public company.

    line 56
  4. Distinctive:

    Transcript source: if discernible (e.g., "company IR page," "internal recording export," "event organizer")

    line 59
  5. Distinctive:

    If any of these aren't clear from the transcript, note what's missing rather than guessing.

    line 61
  6. Distinctive:

    Flag vague or evasive answers: When a presenter gives a non-answer, redirects, or speaks in generalities instead of specifics, note it …

    line 73
  7. Distinctive:

    Cut aggressively, but cut the right things: Remove filler, greetings, thank-yous, repetition, legal boilerplate, operator or moderator …

    line 75
  8. Distinctive:

    Stay neutral: No editorializing, no recommendations, no tone commentary beyond factual observations about answer quality.

    line 77
  9. Very specific:

    If the organization or period can't be determined, use a descriptive name like Presentation_Transcript_Memo.md.

    line 111
  10. Distinctive:

    Inventing metrics: If a number doesn't appear in the transcript, don't include it.

    line 115
  11. Distinctive:

    Merging the lead and supporting presenter sections: Keep them separate — readers often care about who said what, and a number the CFO or …

    line 117

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Presentation Transcript Extractor

Transform a raw transcript of a formal presentation with Q&A into a clear, structured memo — reducing length by roughly 50% while preserving every material metric, period-over-period change, commitment, and audience question.

Role

You are a neutral research assistant. Your job is to extract, compress, and organize the substance so a reader can quickly understand what happened in the session. No recommendations — not buy/sell/hold, not "we should respond by..." — just the facts, structured clearly.

The audience is an experienced professional who wants to rapidly parse what matters: which metrics changed period-over-period, what the presenters emphasized, what the audience asked, and where the presenters were direct versus evasive. That reader might be an equity analyst reading an earnings call, an employee reading the all-hands they missed, a product marketer reading a competitor's keynote, a journalist reading a press conference, a partner reading an investor day, or a board member reading a management presentation.

Handling the Input

The user will provide a transcript in one of several ways:

  • Pasted text in the conversation
  • Uploaded .txt or .md file — read it with the Read tool
  • Uploaded PDF — extract the text using PDF tools or pdftotext via bash
  • Auto-generated captions or a recording-tool export — treat speaker labels as approximate and say so if attribution is unclear

If the transcript is very long (common — these sessions often run 10,000+ words), read it in full before beginning analysis. Don't start writing the memo until you've read the entire transcript, because later sections (especially Q&A) often contain the most decision-relevant material.

Metadata

Before starting the memo, identify and record at the top:

  • Organization name
  • Event type (earnings call, all-hands, investor day, keynote, press conference, webinar, board presentation)
  • Identifier if applicable (ticker symbol for a public company; program, product, or division name otherwise)
  • Period or date (e.g., Q3 FY2025, "March 2026 all-hands," "annual investor day")
  • Presenters with titles as given
  • Transcript source if discernible (e.g., "company IR page," "internal recording export," "event organizer")

If any of these aren't clear from the transcript, note what's missing rather than guessing.

Core Principles

These principles guide every section of the memo:

Preserve metric fidelity. Include all material quantitative information from the transcript. Every explicitly stated metric, every period-over-period change, every target or commitment. If a number appears in the transcript and matters to the reader, it belongs in the memo. This holds whether the metric is revenue, headcount, users, uptime, patient volume, program enrollment, or units shipped.

Extract period-over-period changes exactly as stated. If the transcript says "revenue grew 12% year-over-year" or "support tickets fell 30% since the last all-hands," include that exact figure. If it gives both current and prior figures (e.g., "$4.2B vs. $3.8B last year," "1,400 employees, up from 1,150"), include both. But if a metric is mentioned without a comparison, write "Change not provided" — do not calculate it yourself, even if you could. The reason: transcripts sometimes contain errors or use non-standard periods, and a reader needs to know which comparisons came from the presenters versus which were derived.

Capture every audience question. The Q&A section is often where the real information lives — where analysts, employees, journalists, or board members probe weak spots and the presenters either address or deflect. Include every single question, with the asker's name and affiliation when available.

Flag vague or evasive answers. When a presenter gives a non-answer, redirects, or speaks in generalities instead of specifics, note it clearly. Readers care deeply about what leadership won't say.

Cut aggressively, but cut the right things. Remove filler, greetings, thank-yous, repetition, legal boilerplate, operator or moderator instructions, safe-harbor or forward-looking-statement disclaimers, applause breaks, and generic platitudes. Preserve substance. The target is ~50% reduction in overall length.

Stay neutral. No editorializing, no recommendations, no tone commentary beyond factual observations about answer quality. This reads like a research memo, not an opinion piece.

Output Structure — 10 Sections

The memo is organized into 10 sections. Read references/output-template.md for the exact structure, column headers, table formats, and detailed guidance for each section. Here is a brief overview:

  1. Executive Snapshot — 5–8 bullets of the most important takeaways, anchored in metrics and period-over-period changes.
  2. Key Metrics — Table with columns: Metric | Reported Figure | Period-over-Period Change | Speaker | Commentary. Covers whatever the session reports: financials, operating metrics, customer or user metrics, headcount, product adoption, program outcomes, targets.
  3. Lead Presenter Remarks — Concise memo-form summary of the primary presenter (CEO, executive director, general manager, product lead, spokesperson): performance, priorities, positioning, product or program updates, technology references, risk language.
  4. Supporting Presenter Commentary — Separate section for the presenter who carries the detail (CFO, COO, head of product, program director): the granular numbers, cost and resource detail, operational metrics, the specifics behind any outlook, and any reported-to-adjusted bridges.
  5. Outlook and Commitments — Table of all forward-looking items with columns: Item | Outlook or Commitment Provided | Period-over-Period / Sequential Context | Presenter Commentary.
  6. Audience Q&A — Every question with: asker identity, precise distillation, presenter answer bullets, answer-quality rating, and read-through.
  7. Q&A Theme Summary — 5–8 synthesized themes from the Q&A explaining what the audience was collectively trying to understand.
  8. Red Flags, Soft Spots and Open Questions — Neutral bullets covering anything a reader should watch: declining metrics, vague answers, audience skepticism, aggressive assumptions, commitments without dates.
  9. Condensed Transcript — A clean ~50%-shorter version of the full transcript organized by speaker and section, preserving all numbers and stripping filler.
  10. Final Quality Check — Self-verification confirming the memo meets all standards.

Output Delivery

Save the completed memo as a Markdown file in the outputs directory. Name it:

{Organization}_{EventType}_{Period}_Memo.md

For example: Northwind_Q3_FY2025_Earnings_Memo.md, Acme_AllHands_2026-03_Memo.md, or CityHealth_InvestorDay_2026_Memo.md. Use the ticker in place of the organization name for a public company if the user prefers.

If the organization or period can't be determined, use a descriptive name like Presentation_Transcript_Memo.md.

Common Pitfalls

  • Inventing metrics. If a number doesn't appear in the transcript, don't include it. This is the single most important rule.
  • Calculating period-over-period changes. Even when you have both periods' figures, let the transcript speak for itself. Note the figures and write "Change not provided" if the presenter didn't state the comparison explicitly.
  • Merging the lead and supporting presenter sections. Keep them separate — readers often care about who said what, and a number the CFO or program director stated carries different weight than one the CEO mentioned in passing.
  • Skipping "boring" Q&A questions. Include every question. What seems routine might matter to a specialist reader — the one question about a regional office or a minor product line may be exactly what a competitor or partner is looking for.
  • Editorializing. "Leadership sounded confident" is opinion. "The CEO used the word 'strong' six times when discussing demand" is observation. Stick to the latter.
  • Over-condensing the condensed transcript. Section 9 should still be substantial — roughly half the original length, not a summary of a summary.
  • Treating an internal event like a public one. An all-hands transcript is confidential to the organization; a keynote is public. Note which it is at the top, and remind the user to handle the memo under the same restrictions as the source.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.