agentleFS
pass-agentpass-agent/loomkin/.agents/skills/vault-meeting/SKILL.md

vault-meeting

Skill179 starsChanged 8 months agoMITView on GitHub

Process a meeting transcript — extract decisions, action items, and key discussion points into structured vault entries

The file

Show its rules and text

We split this skill into the rules it gives an agent

13 rules. About 0 are common, 8 are distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈8

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    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 (60+ minutes of conversation, or the user requests team processing), spawn a team to parallelize the work:

    line 23
  2. Distinctive:

    Pasted directly: Use the text from the user's message

    line 52
  3. Very specific:

    Local file reference: Use vault_read or file_read as appropriate

    line 55
  4. Very specific:

    If the user says something like "process the meeting from Drive" without a specific file, use ask_user to get the file ID or link.

    line 57
  5. Very specific:

    Search for a prep file matching the meeting date: vault_search(query: "prep", entry_type: "meeting", tags: ["prep"])

    line 61
  6. Distinctive:

    Read it to understand the planned agenda

    line 65
  7. Very specific:

    If the transcript is very long (appears to be 20+ minutes of conversation), offload it to a context keeper: context_offload(topic: …

    line 71
  8. Distinctive:

    Read through and identify:

    line 78
  9. Distinctive:

    Follow-ups: Items needing attention even without explicit assignment

    line 97
  10. Distinctive:

    When redacting, think holistically about context.

    line 120
  11. Distinctive:

    Atomic notes: If discussion surfaced a reusable concept or strategy, create it as a note and link to the parent topic

    line 130
  12. Distinctive:

    If a prep file existed:

    line 134
  13. Very specific:

    Add processed: true to frontmatter

    line 137

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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

Process a meeting transcript and extract structured information into the knowledge base.

Team Mode (Large Transcripts)

If the transcript is very long (60+ minutes of conversation, or the user requests team processing), spawn a team to parallelize the work:

team_spawn(
  team_name: "meeting-processing",
  purpose: "Process a long meeting transcript into structured vault entries",
  roles: [
    %{name: "meeting-lead", role: "lead"},
    %{name: "vault-researcher", role: "researcher"},
    %{name: "vault-writer", role: "coder"},
    %{name: "vault-reviewer", role: "reviewer"}
  ]
)

Team roles:

  • meeting-lead: Orchestrates extraction, manages redaction judgment, coordinates. Reads the transcript, identifies decisions/action items/topics, delegates creation to the writer.
  • vault-researcher: Queries vault for related context — prior decisions on discussed topics, open tasks for mentioned projects, recent checkins from attendees. Provides context to the lead and writer.
  • vault-writer: Creates the meeting note, decision records, atomic notes, and kanban items with proper formatting and linking. Works from the lead's extraction.
  • vault-reviewer: Validates output quality — checks for temporal language in notes, verifies link targets exist, ensures frontmatter is complete. Runs vault_audit on created entries.

The researcher and writer can work in parallel on different aspects. For shorter meetings (under 60 minutes), skip team mode and process single-agent.


Step 0: Get the Transcript

Determine where the transcript is:

  • Pasted directly: Use the text from the user's message
  • Google Drive link or file ID: fetch_content(source: "google_drive", identifier: "{file_id}")
  • URL: fetch_content(source: "url", identifier: "{url}")
  • Local file reference: Use vault_read or file_read as appropriate

If the user says something like "process the meeting from Drive" without a specific file, use ask_user to get the file ID or link.

Step 1: Check for Prep

Search for a prep file matching the meeting date: vault_search(query: "prep", entry_type: "meeting", tags: ["prep"])

If prep exists:

  • Read it to understand the planned agenda
  • Track which topics get covered during processing
  • You will mark covered topics and report coverage at the end

Step 2: Handle Long Transcripts

If the transcript is very long (appears to be 20+ minutes of conversation), offload it to a context keeper: context_offload(topic: "meeting-transcript-{date}", content: ...)

This preserves the full transcript at high fidelity without consuming your context window.

Step 3: Analyze the Transcript

Read through and identify:

Attendees — who spoke in the meeting

Decisions — commitments to a course of action. Look for:

  • "Let's go with...", "We'll do...", "We decided..."
  • Choosing between alternatives
  • Agreeing on a direction

For each decision, assess:

  • Who proposed it (who said "I think we should..." or "What if we...")
  • Who decided (who gave final approval — "Sounds good", "Let's do it")
  • Scope: company | product | project
  • Reversibility: one-way (hard to undo — hiring, equity, pivots) | two-way (easy to change — features, tools)

Action items — be thorough. Look for ALL of these patterns:

  • Explicit: "[Person] will...", "Can you...", "Take care of..."
  • Volunteering: "I'll handle that", "Let me do...", "I can take..."
  • Implied: If someone says they'll improve/fix/create something, that is a task
  • Follow-ups: Items needing attention even without explicit assignment

Key discussion points — topics that got meaningful airtime

Summary — 2-3 sentence overview

Step 4: Redaction Judgment

This is a shared company vault. Apply judgment about what belongs in shared records.

Auto-redact (do it, no confirmation needed):

  • Phone numbers, addresses, SSNs
  • Financial figures (salaries, investment amounts)
  • Passwords, API keys

Flag for user confirmation (use ask_user):

  • Explicit removal requests ("off the record", "don't add that")
  • Personal asides clearly unrelated to work

Omit entirely (don't even flag):

  • HR/personnel matters — note "Personnel discussion - details in private records"
  • Individual criticism of team members

When redacting, think holistically about context. A single-line gap surrounded by reactions is worse than no redaction — remove the full exchange.

Step 5: Create Entries

  1. Meeting note: vault_create_entry(entry_type: "meeting", ...) with full extracted content
  2. Decision records: For each decision:
    • vault_create_entry(entry_type: "decision", ...) — creates DR-YYYY-NNN automatically
    • decision_log(node_type: "decision", ...) — adds to the live decision graph with confidence score
    • vault_link(source_path: meeting, target_path: decision, link_type: "decides")
  3. Action items: vault_kanban(action: "add", ...) for each task, linked to the meeting
  4. Atomic notes: If discussion surfaced a reusable concept or strategy, create it as a note and link to the parent topic

Step 6: Update Prep (if exists)

If a prep file existed:

  • vault_update_entry to check off covered topics (- [ ] to - [x])
  • Count coverage: "Discussed X/Y agenda topics"
  • Add processed: true to frontmatter

Step 7: Refresh Dashboard

vault_dashboard(dashboard_type: "activity") then use the result to update the index entry.

Step 8: Report

Summarize what was created:

  • Meeting note path
  • Decision records created (list with DR numbers)
  • Action items added (count, grouped by assignee)
  • Prep coverage (if applicable)
  • Any flagged redactions that need review

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.