Claude Skills Companion

This book is static. The Claude skills that ship with it are how it becomes an active learning loop.

This appendix documents the six skills, their invocation patterns, the data they read/write, and how they work together.

Install

From the repo root:

./scripts/install-skills.sh

This symlinks the three MVP skills (si-daily-challenge, si-post-game, si-at-the-table) into ~/.claude/skills/, so Claude Code recognizes them by name.

Uninstall:

./scripts/install-skills.sh --uninstall

The six skills

MVP (shipped now)

si-daily-challenge — today’s drill

When: “daily Spirit Island challenge”, “SI drill today”, /si-daily-challenge.

What it does: picks today’s spirit × adversary × level × learning goal based on data/playlog.csv. Anti-repeats recent spirits; progresses difficulty by 2-wins-in-a-row.

Output: a compact markdown block with the challenge, links to chapters to read first, a “write on paper: T3 milestone” prompt.

Depends on: data/playlog.csv, data/spirits.json, data/adversaries.json.

si-post-game — log + reflect

When: “SI post-game”, “log my game”, any sentence describing a just-finished game.

What it does: normalizes Brett’s shorthand into a CSV row, appends to data/playlog.csv, then surfaces 1–2 likely learning points by diffing his notes against the spirit chapter’s Common Mistakes.

Output: logged-confirmation line, 1–2 takeaways with chapter links, rolling stats, next-session suggestion.

Depends on: data/playlog.csv, spirit chapters.

si-at-the-table — live tactical lookup

When: Brett mid-game describing a board state + asking a question.

What it does: reads the spirit + adversary chapter, returns 2–3 options with tradeoffs. Never solves the turn.

Output: terse options-with-tradeoffs block. Designed to be read in 10 seconds at the table.

Hard rule: anti-alpha. Never prescribes specific plays.

Depends on: chapter content.

Deferred (M2–M4)

si-combo-drill — 2-spirit pairing drill

Status: skeleton stub only; full logic lands with Part VI in M3.

si-matchup-drill — adversary-focused drill

Status: skeleton stub only; full logic lands with the adversary chapters in M2.

si-aspect-explorer — aspect recommendation

Status: skeleton stub only; full logic lands with spirits/aspects.md in M4.

The loop

graph LR
  A[si-daily-challenge] --> B[Read chapter 15min]
  B --> C[Play the challenge]
  C --> D[si-post-game]
  D --> A
  C -.->|mid-game| E[si-at-the-table]

1. Morning: si-daily-challenge picks today’s drill.
2. Pre-game: 15 min with the spirit + adversary chapter.
3. Play.
4. Mid-game if needed: si-at-the-table for tactical signal.
5. Post-game: si-post-game logs + surfaces 1–2 learnings.
6. Tomorrow’s challenge shifts based on today’s result.

After ~20–30 loops, you’ll notice you’re reaching for chapter less — patterns are in your head.

The living-document feedback path

Running the skills changes the data, which changes the book:

• data/playlog.csv grows. The Self-Tracking chapter teaches you to read your own data.
• si-post-game outputs become revision tickets. If a chapter’s common-mistake list doesn’t predict what actually cost you games, the chapter needs updating.
• New chapters get added as your ladder progresses — stub → full as you learn that spirit.

Data schema

data/playlog.csv

date,player_count,spirits,aspects,adversary,level,scenario,board,result,rounds,fear_level_end,blight_end,notes

• date: ISO YYYY-MM-DD
• player_count: integer
• spirits: comma-separated slug list (for multi-spirit games)
• aspects: aspect names, comma-separated
• adversary: slug from data/adversaries.json
• level: integer 0–6
• scenario: slug from data/scenarios.json or empty
• board: letter A–F, letter-list, or empty
• result: win | loss
• rounds: integer
• fear_level_end: integer (fear pool count at game end)
• blight_end: integer
• notes: freeform, Brett’s verbatim commentary

data/spirits.json

Schema at file top. Key fields used by skills:

• brett_priority: prioritize in daily-challenge rotation
• chapter_status: full vs stub; only full chapters are linked as “read first”
• archetypes: used by si-combo-drill once shipped

data/adversaries.json

Per-adversary: slug, name, expansion, difficulty_levels, pressure_shape, chapter_path, chapter_status.

data/scenarios.json

Per-scenario: slug, name, expansion, difficulty_mod, pressure_shape, chapter_path.

Extending the skills

The skills are plain markdown with YAML frontmatter (per Claude’s writing-skills conventions). To add a new skill or change behavior:

1. Edit skills/{skill}/SKILL.md.
2. Re-run ./scripts/install-skills.sh --force to refresh symlinks.

Privacy

All skills write to data/playlog.csv locally. Nothing is sent anywhere except what Claude Code normally sends when you invoke a skill in-session.

Last revised: 2026-04-19