Imported from imisic/claude-obsidian-vault (.claude/skills/w-1on1/SKILL.md). Install upstream with npx skills add imisic/claude-obsidian-vault --skill w-1on1. Copyright stays with the author.

1on1 Prep

Prepare for a 1on1 meeting with $ARGUMENTS.

Phase 0: Build indexes (< 2 seconds)

Run these two scripts in parallel:

python _scripts/build-person-index.py
python _scripts/build-open-actions.py

Then extract targeted data for this person:

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.