agentsclimarketplace

Transcript analysis

Skill crystal-project-inc/crystal-knows-plugin/skills/transcript-analysis

Analyze a call/meeting transcript by each participant's DISC type—resolve attendees' DISC via Crystal, then explain what landed, what didn't, and why. Use to debrief or review a call.From its SKILL.md

Install
npx -y skills add crystal-project-inc/crystal-knows-plugin --skill transcript-analysis

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

7.5 KB, as published. Nobody here has run it

Transcript Analysis (DISC-grounded)

Analyze a call/meeting transcript and explain what went well and why, and what went poorly and why, grounded in each participant's DISC personality type. Personality data comes from the Crystal MCP server — never guess a person's DISC type; resolve it with the tools below.

When to use

Trigger when the user wants to debrief, review, or get feedback on a conversation: "here's my call transcript", "why did this meeting go badly?", "review this sales call", "how did I do on this demo?", etc.

Prerequisites

This skill depends on the Crystal MCP tools being available. In claude.ai chat, enable the Crystal connector (Settings → Connectors); in Claude Code, the crystal-knows plugin provides them. It uses these tools:

  • multi_search — resolve up to 25 people at once by email or LinkedIn URL. Returns per person: disc_type, disc_archetype, traits, name, id.
  • search — single lookup that also supports name-only queries among profiles that already exist in Crystal. Use as a fallback when you only have a name, not an email/LinkedIn URL.
  • generate_prompt / selling_to — optional depth if the user asks for follow-up guidance or a sales angle after the analysis.

If the tools aren't available, tell the user to install and authenticate the crystal-knows plugin (OAuth) first, then stop.

Procedure

1. Get the transcript

If the user hasn't already pasted a transcript, prompt for it explicitly:

Paste the full call/meeting transcript below. Include speaker labels/names if you have them — that's how I'll map lines to each person.

Do not proceed without a transcript. If it's very long, work from the whole thing; do not truncate silently.

2. Identify participants

Extract the list of distinct speakers/participants from the transcript (speaker labels, names, "me"/"rep" vs. "client", etc.). Present the list back to the user so they can confirm and correct it.

3. Resolve DISC types (the critical step)

multi_search matches on email or LinkedIn URL only — it does not accept names. So:

  1. For each participant, gather an email or LinkedIn URL. Ask the user to fill in any you don't have:

    To pull DISC types I need an email or LinkedIn URL for each person. I have:

    • Jane Doe → (need email or LinkedIn)
    • John Smith → [email protected] Please fill in the missing ones (or say "skip" for anyone).
  2. Call multi_search with the collected emails/LinkedIn URLs (batch, max 25).

  3. For anyone you only have a name for, fall back to the search tool with the name — it can resolve profiles that already exist in Crystal.

  4. Note anyone who couldn't be resolved (in multi_search's failed array or a search miss). Continue with whoever resolved; call out the unresolved people so the user knows their analysis is DISC-blind.

Never invent a DISC type. If a person can't be resolved and the user can't supply an identifier, analyze their contributions descriptively but label the DISC read as "unknown."

4. Establish a citation scheme

Every point in the analysis must cite the exact moment in the transcript it rests on. Choose the citation format by what the transcript already provides, in this priority order:

  1. Timestamps — if lines carry timestamps (e.g. [00:03:12], 12:04, (1:15)), cite with the timestamp of the relevant line, e.g. [00:03:12].
  2. Explicit line numbers — if the transcript is already numbered (e.g. 14: or L14 prefixes), cite that number, e.g. [L14].
  3. Counted line numbers — if neither is present, number the non-empty content lines yourself, starting at 1 (each speaker turn / line = 1), and cite those, e.g. [L14]. Add a one-line note in the output that line numbers were auto-assigned by counting.

Rules:

  • Pick one scheme (whichever the priority selects) and use it consistently throughout.
  • Point at the specific line(s) the claim rests on; cite a short range when a point spans turns, e.g. [L14–15] or [00:03:12–00:03:40].
  • Every "what landed / what didn't" bullet carries a citation. No citation, no claim.

5. Analyze the transcript against DISC

For the conversation as a whole and for key moments, connect observed behavior in the transcript to each participant's DISC type. Use the traits and disc_archetype returned by Crystal as your evidence base, plus the DISC reference below.

Assess, per participant and for the interaction overall:

  • What landed well, and why — moments where communication matched the other person's DISC preferences (e.g. giving a D the bottom line up front; giving a C data and specifics; giving an S reassurance and a steady pace; giving an I energy and social connection).
  • What fell flat, and why — DISC mismatches (e.g. burying the ask in small talk with a D; rushing a S; hand-waving specifics with a C; being cold/transactional with an I). Cite the transcript line (per the citation scheme) that shows it.
  • What to do differently next time — concrete, DISC-specific adjustments.

Always tie claims to evidence: cite the transcript (per the citation scheme), quote the relevant snippet, then give the DISC reason. Avoid generic advice that isn't anchored in both the transcript and a resolved DISC type.

6. Output

Produce a structured debrief. Prefix every evidence bullet with its citation ([00:03:12] for timestamps, [L14] for line numbers):

## Participants & DISC
- <Name> — <DISC type> (<archetype>): <one-line read>
- <Name> — unknown (couldn't resolve)

_(Line numbers auto-assigned by counting — include this note only when the
transcript had neither timestamps nor explicit line numbers.)_

## What went well
- `[L12]` "<quoted snippet>" → <why it worked, in DISC terms>

## What didn't land
- `[L18]` "<quoted snippet>" → <why it missed, in DISC terms>

## Do differently next time
- `[L18]` <person/moment> → <DISC-specific adjustment>

Keep it tight and evidence-led. If the user asks for a follow-up email or a sales angle afterward, hand off to revise_email, generate_prompt, or selling_to.

DISC quick reference

Use as interpretation scaffolding — the authoritative traits come from Crystal's traits / disc_archetype per person.

  • D — Dominance (direct, results-driven, fast, impatient with detail): wants the bottom line, options, and control. Lands well: brevity, confidence, outcomes. Falls flat: long preamble, hedging, excessive detail.
  • I — Influence (social, enthusiastic, optimistic, relationship-first): wants energy, rapport, big-picture vision. Lands well: warmth, stories, collaboration. Falls flat: cold/transactional tone, drowning them in data.
  • S — Steadiness (patient, supportive, dislikes pressure and sudden change): wants reassurance, a steady pace, and clear next steps. Lands well: empathy, no rush, stability. Falls flat: high pressure, rushing, abrupt shifts.
  • C — Conscientiousness (analytical, precise, skeptical, detail-oriented): wants accuracy, evidence, and logical structure. Lands well: specifics, data, precision. Falls flat: vagueness, hype, unsupported claims.

Two-letter types (e.g. Di, Sc) blend the primary and secondary — weight the primary letter and soften with the secondary.

Keep looking

Skills are one crate of 608,475. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.