Skill: Kickstart Backfill

⚠️ Special skill — single use

This skill is meant to be invoked once, on Day 1 of the system. It takes time (30–60 min depending on the window) and consumes context. It's the initial investment to start with an already-living vault instead of from zero.

NEVER invoke it as a scheduled task.

When to activate

• The user explicitly says: "kickstart", "backfill", "initialize the vault", "pre-fill with [period]".
• The user mentions wanting to start the second brain with an existing state of the world.

Preflight

1. Read _CLAUDE.md and 00-inbox/MY-PROFILE.md.

2. Ask the user for 3 inputs:

   • Time window: 1 month? 3 months? 6 months? (recommendation: start with 3 months).
   • Sources to scan: email only? + calendar? + drive? + chat? (recommendation: calendar + email first).
   • Detail level: exhaustive (all interactions) or strategic (top stakeholders + active projects only)? (recommendation: strategic for the first pass).

3. Confirm the scope before starting — show an estimate ("I'll scan ~N calendar events, ~N email threads, and create about N notes. ETA: ~X min. Go?").

General strategy

⚠️ Not a raw dump. The goal is NOT to create 500 low-level notes. It's to capture the user's existing mental model:

• Who are their N regular stakeholders?
• What are their N active projects?
• What are the N major decisions made recently?
• What are the N significant meetings?

Aim for 30–60 notes total, not 500. Quality > volume.

Process in 5 phases

Phase 1 — Stakeholder mapping (15 min)

1.1 Extraction from calendar

• Scan all events in the window.
• Count frequency per participant (excluding the user).
• Top 20 by frequency = candidates for person notes.

1.2 Extraction from email

• Scan the threads in the window.
• Count threads per sender (filter out noreply / newsletters).
• Cross-reference with the calendar list.

1.3 Presentation and validation

Show the user a proposed shortlist:

I identified 23 recurring people over the last 3 months.
Here are the top 15, ranked by interaction frequency:

1. Alex Rivera — 32 calendar events + 47 email threads — probably [direct-report]
2. [Name] — 18 events + 25 threads — probably [peer or manager]
3. ...

For each, I'll create a note in 02-people/ with:
- Filled frontmatter (inferred role, region, relationship)
- Timeline pre-filled with major meetings and threads
- Compiled truth ENRICHED only with what's observable (no speculation)

Do you want to:
A) Proceed on the 15?
B) Adjust the list (add/remove)?
C) More restrictive mode (top 10 strategic only)?

⚠️ Ask for validation before creating. This phase is critical: it's the base of the strategic CRM.

1.4 Create the person notes

For each validated person:

Frontmatter: fill with what's inferable

• role: from email signatures or calendar descriptions.
• team, region: best guess + flag if uncertain.
• relationship: ask the user if ambiguous.
• last-interaction: date of the last event/thread.
• date (first logged interaction): the oldest in the window.

Compiled truth: stay minimal. Include ONLY:

• Observable professional background (role, team, trajectory if visible in signatures/profiles).
• Explicit marker: "Compiled truth to be enriched through future interactions".

Do NOT include:

• Psychological inferences.
• Working style (not enough reliable historical signal).
• Motivations (same).

Timeline: create entries grouped by month for the backfilled window:

## Timeline

### 2026-03 (Backfilled — aggregated)
- **Meetings** (8): 5 team standups, 2 1-1s, 1 stakeholder review
  - Recurring topics: qualification script V2, training
  - See: [[04-meetings/2026-03-15-1to1-alex]], [[04-meetings/2026-03-22-...]]
- **Email threads** (12): mostly about [topic]
- **Observation**: *(backfilled, based on calendar/email only, to be enriched)*

### 2026-04 (Backfilled — aggregated)
...

⚠️ Explicit (Backfilled) marker on each entry so future-Claude knows these are not real-time observations but a post-hoc reconstruction.

Phase 2 — Project mapping (10 min)

2.1 Detect candidate projects

Scan:

• Recurring topics in email threads (3+ threads on the same topic = candidate project).
• Recurring calendar events (e.g. weekly project sync).
• Drive folders with recent activity.
• Active chat channels (if connected).

2.2 Validate with the user

Candidate projects detected:
1. "Qualification Script V2" — 12 events, 23 emails, active drive folder → STRONG
2. "Direct Report Coaching" — weekly recurrence with Alex → STRONG
3. "Process Unification" — 8 emails, 3 drive docs → MODERATE
4. "Tooling Migration" — a few mentions → WEAK

Which do you validate? Want to add any not detected automatically?

2.3 Create the project notes

For each validated project, create 03-projects/<slug>.md with:

• Frontmatter: status: active (default), ask the user for priority.
• Goal: ask the user (one sentence).
• Stakeholders: auto-detected from calendar/email.
• Timeline: key events detected in the window, grouped by month.
• Explicit (Backfilled) marker on retroactive timeline entries.

⚠️ Ask the user for each project: "Goal in one sentence?" — that's what turns a descriptive note into a strategic one.

Phase 3 — Historical decisions (10 min, optional)

3.1 Detection

Scan emails and drive docs for keywords signaling decisions:

• "we've decided", "approved", "let's go with", "final decision".
• "reversal", "we're changing course", "stopping".
• Threads with "decision" in the subject.

3.2 Validation

List candidate decisions to the user:

I detected 7 potential decisions in the last 3 months:

1. (2026-03-08) Centralize tier-1 in one hub — source: email thread [...] → major decision?
2. (2026-04-12) Onboarding training redesign — source: meeting [...]
...

Which should I log in 05-decisions/?

3.3 Creation

For validated decisions, create 05-decisions/YYYY-MM-DD-<slug>.md with:

• Frontmatter status: implemented (default, since retroactive).
• Decision, Context, Rationale sections (from email/meeting sources).
• (Backfilled) marker.
• Important note: "Logged retroactively on YYYY-MM-DD via kickstart-backfill. Source materials reviewed but full rationale may be incomplete."

Phase 4 — Major meetings (10 min, optional)

For the meetings the user identifies as strategically important (max 10):

• Create notes in 04-meetings/.
• If a transcript is available → ingest via the standard meeting-ingest process.
• Otherwise → minimal note with participants, topic, known outcome, marker (Backfilled, no transcript).

Anti-pattern: don't create 50 meeting notes for the window. Only the key meetings (stakeholder reviews, manager 1-1s, major decisions).

Phase 5 — Knowledge baseline (5 min, optional)

If the user allows, create 1–3 06-knowledge/ notes to crystallize emerging patterns:

• Observed anti-patterns.
• An applied framework.
• Major lessons learned detected.

⚠️ Ask for validation for each knowledge note. This is interpretive content and must be validated by the user.

Generate the backfill report

At the end, create a meta note: 06-knowledge/kickstart-backfill-YYYY-MM-DD.md

---
date: YYYY-MM-DD
type: backfill-report
tags: [meta, backfill]
ai-first: true
---

## For future Claude

Report of the kickstart backfill executed on [date]. Window covered: [start-date] →
[end-date]. This documents what was reconstructed retroactively vs. what was captured
in real-time afterward.

## Method

- Time window: [...]
- Sources scanned: [email | calendar | drive | chat]
- Mode: strategic | exhaustive

## Stats

- People created: N
- Projects created: N
- Decisions logged: N
- Meetings created: N
- Knowledge notes seeded: N

## Important caveats for future-Claude

All notes created in this operation are marked `(Backfilled)` in their timeline.
Behavioral observations and compiled truths are DELIBERATELY minimal — based only
on the observable (role, interaction frequency), not the interpretive.

**When using challenge-decision with this vault in the first weeks**: weight
backfilled notes lower than real-time notes. A backfilled note = ~30% of the weight
of a real-time note.

As the user works the system in real time, the notes enrich with real observations.
After ~6–8 weeks, the "backfill" effect is diluted.

## Links to created notes

[Auto-generated list of notes]

Final report to the user

✓ Kickstart backfill complete
✓ Window: 2026-02-20 → 2026-05-20 (3 months)

Created:
- 14 people notes (02-people/) — top stakeholders identified
- 4 project notes (03-projects/) — all status: active
- 3 historical decisions (05-decisions/)
- 6 key meetings (04-meetings/)
- 1 knowledge note (observed anti-patterns — to validate)
- 1 meta report (06-knowledge/kickstart-backfill-2026-05-20.md)

⚠️ To do within the next 7 days to finalize:
1. Re-read the 14 people notes, complete the "Compiled truth" with your real mental model
2. Validate the "Goals" of the 4 projects (I put best-guesses to confirm)
3. Check that no people note is a duplicate (fuzzy match is limited in this mode)

The vault is now seeded. Day 1 operational.
Your first daily-brief will have material to cross-reference.

Anti-patterns to avoid

❌ Over-inferring — no psychology, no invented motivations. Stick to observable. ❌ Too many notes — 30–60 max total. Not 500. ❌ Forgetting the (Backfilled) marker — critical so future-Claude weights correctly. ❌ Skipping user validation — each phase has its validation gate. ❌ Trying to resolve everything in one pass — it's OK to leave stubs to enrich. ❌ Including speculative decisions — if a decision isn't explicitly documented, don't log it.

Alternative modes

"Light" mode (15 min)

For a faster start:

• Skip phases 3 and 4.
• Limit phase 1 to top 10 people.
• Limit phase 2 to top 3 projects.

"Heavy" mode (60–90 min)

For a very rich start (vault at 100+ notes):

• Extend the window to 6 months.
• Include all stakeholder meetings.
• Create knowledge notes per project (retroactive mini-PRD).

"Topic-focused" mode

To start focused on one specific project:

• Skip generic people.
• Deep-dive one project: all meetings, all decisions, all involved stakeholders.
• Output: an ultra-documented project from Day 1.

Ask the user for their mode at preflight.

Once the skill is done

This skill is archivable. Once used:

• You can disable the skill (move it out of .claude/skills/ for clarity).
• Or keep it for reference (in case the user wants to re-backfill an older period later).

Typical invocation: once on Day 1, sometimes a second time after 6 months to ingest a missed period (if there was a gap or the user wants to extend retroactively).