Skip to content
Back to skills

W 1on1

ASecurity

Prepare for a 1on1 meeting with a specific person. Gathers previous meeting history, open action items, and creates a pre-populated meeting note.

  • 3 stars
  • 0 votes
  • 0 copies
  • 3 views
  • Added September 19, 2026
ai-agentspythonbash

Security analysis

A100/100

Scanned September 19, 2026

npx -y skills add imisic/claude-obsidian-vault --skill w-1on1 --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of W 1on1?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for W 1on1
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/imisic-w-1on1/badge)](https://www.skillsdirectory.com/skills/imisic-w-1on1)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: w-1on1
description: Prepare for a 1on1 meeting with a specific person. Gathers previous meeting history, open action items, and creates a pre-populated meeting note.
user-invocable: true
argument-hint: "[Person Name]"
---

# 1on1 Prep

Prepare for a 1on1 meeting with $ARGUMENTS.

## Phase 0: Build indexes (< 2 seconds)

Run these two scripts in parallel:
```bash
python _scripts/build-person-index.py
python _scripts/build-open-actions.py
```

Then extract targeted data for this person:
```bash
python _scripts/prep-1on1-data.py --person "$ARGUMENTS"
```

The script uses **differential windowing** by default: it anchors to the last 1on1 date as a cutoff, so only new activity since the last meeting surfaces. If no prior 1on1 exists, it falls back to `--days 21`.

If the user explicitly requests a custom window (e.g., `/w-1on1 Bruno --days 30`), pass `--days N` to override.

Capture the JSON output. This is all the structured data the agent needs, including `carry_forward` items from the last 1on1's "Next time" section.

## Phase 1: Invoke 1on1-prep agent (pin to Opus)

Dispatch 1on1-prep as a `general-purpose` Agent with **`model: "opus"`** set explicitly on the Agent call, then have it invoke the `1on1-prep` skill via the Skill tool. The skill's frontmatter `model:` is NOT honored under `subagent_type: "general-purpose"` dispatch, so without the explicit parameter this prep silently inherits the session model instead of Opus.

Pass the JSON output from prep-1on1-data.py to the agent as context.
The agent should NOT search for interactions or grep for action items: all structured data is in the JSON.

The agent's job is now synthesis only:
1. Read the person's file in 04-People/ for role context
2. Read the prep JSON data (passed as context)
3. If past 1on1s exist in the JSON, read the last 1-2 for discussion detail (only if summaries are insufficient)
4. Synthesize talking points: carry-forward items first, then new interactions since last 1on1, then open actions
5. Create the meeting prep note in `00-Inbox/`

Report: new note path, open items, suggested topics.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.