1on1-prep

1:1 Prep & Capture

Prep reads the knowledge graph to surface context, commitments, and signal. Capture writes verbatim observations with deterministic tags.

Announce: "I'm using the 1on1-prep skill to help you prepare for your 1:1."

Flow: Prerequisites → Invocation → Person Lookup → Bootstrap → Phase Detection → Mode Detection → Prep or Capture.

Reference files (read on demand, not upfront):

• graph-schema.md — observation format, tags, relations, examples
• pending-sync.md — file format, parsing, error handling
• capture-form.md — 6-prompt form, parsing rules, resolution flow
• questions.md — question bank
• stakeholder-map graph-schema — extended schema (power, category, coverage, advice) shared when stakeholder-map is in use

Prerequisites

Verify memory MCP: mcp__memory__read_graph. If unavailable, warn and set a session-wide flag so ALL writes route to pending-sync for the remainder (not just the current write). Check for existing pending-sync files.

Invocation

/1on1-prep <person-name> [--mode=intake|coaching] [--phase=prep|capture] [--context "..."] [--sync]

--context writes a [context] observation and exits. --sync drains pending-sync files (see pending-sync.md).

Person Lookup

mcp__memory__search_nodes({ query: "<person-name>" }) — exact match → use it, one substring → use it, multiple → ask user, not found → Bootstrap, tool error → stop (prevents duplicate creation).

Bootstrap (New Person)

Requires memory MCP. Four-prompt form: 1) Full name 2) Role/team 3) Background 4) Reports to (optional). Answer #1 = entity name. Others → [context] observations. Check name collision. Only create reports_to if manager exists.

Phase Detection

If no --phase, query calendar (-4h to +24h). Upcoming → Prep. Ended ≤4h → Capture. Multiple → ask. None/unavailable → ask.

Mode Detection

Priority: --mode flag → explicit [mode:*] marker in graph → auto-detect. Intake if <3 [1on1] observations, no [context], no reports_to. Else Coaching. Note: open_nodes doesn't return relations — use search_nodes to check reports_to. Offer graduation nudge if heuristic says coaching but no marker exists.

Prep Phase (Read-Only)

Read Person's full node. Render in order, omit empty sections:

1. Header — ## <Name> [MODE] 1:1 #N with meeting time
2. Context — [context] observations as bullets
3. Open Commitments — unresolved [commitment], oldest first, cap 10
4. Open Follow-ups — unresolved [followup], cap 10
5. Recent Signal — last 2-3 sessions by strategic tag
6. What Others Said — cross-entity search_nodes (cap 5, never read_graph)
7. Suggested Questions — 3-4 from questions.md, 2+ categories

Capture Phase

Before showing the form, check if user indicates no meeting occurred (e.g., "nothing happened", "we didn't meet", "they cancelled"). If so, skip to noshow: write [YYYY-MM-DD][1on1][<mode>][noshow] No capture recorded. Failed → pending-sync.

Otherwise, read capture-form.md for form, parsing, and resolution flow. Tags assigned deterministically by prompt bucket. Write one-at-a-time, failed writes → pending-sync.

Common Mistakes

• Summarizing capture observations instead of recording verbatim — the graph is only useful if entries preserve what was said; paraphrasing loses signal and invents details.
• Auto-resolving open commitments or follow-ups without asking — always confirm with the user before marking prior items resolved; Prep shows them, it doesn't close them.
• Assigning tags by LLM judgment instead of prompt bucket — capture tags are deterministic by form prompt; don't re-tag based on content interpretation.
• Skipping the cancelled/no-show path — if the user indicates no meeting occurred, write the [noshow] entry instead of showing the capture form.
• Reading the full graph when person lookup suffices — use search_nodes for cross-entity lookups; avoid read_graph which pulls the entire corpus into context.