Menü
Context preservation - route learnings to cal.md (permanent) or memories/ (ephemeral)
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Basierend auf der SOC-Berufsklassifikation
SwiftUI iOS/macOS 27 delta reference (WWDC 2026) — Apple-authored references vendored from Xcode 27's exported agent skills, plus Cal SDK-verified extras. Use for any SwiftUI work targeting SDK 27. Use when a SwiftUI view using @State fails to compile with "used before being initialized", "invalid redeclaration of synthesized property", or "extraneous argument label" errors after an SDK update (@State became a macro in SDK 27 — the obvious fix of reordering init assignments is WRONG; consult references/state-macro.md first); when @ViewBuilder/@ContentBuilder code hits ambiguous overloads in overlay/background or type-check regressions; when adding drag-to-reorder to any container via reorderable()/reorderContainer; when working with AsyncImage caching or AsyncImage(request:); when adding swipe actions to rows outside List via swipeActionsContainer(); when controlling toolbar overflow (visibilityPriority, ToolbarOverflowMenu, toolbarMinimizeBehavior); when presenting alert/confirmationDialog from an optional i
iOS/macOS/iPadOS 27 framework reference (WWDC 2026). Corrects stale model knowledge about App Intents, App Schemas, Foundation Models, and Siri AI. Invoke when building or reviewing iOS 27 / WWDC 2026 features, or when a model assumes iOS 26 / SiriKit is current.
Apple Liquid Glass (iOS 26, refined in 27) design system reference. Invoke when building UI, reviewing visual design, or implementing glass effects, background extension, tab bar minimize, typography, spacing, animations, or accessibility.
Advance the pipeline - find and execute next step
Project setup - scan codebase, create Cal structure, generate or improve CLAUDE.md
Scan codebase for small code hygiene wins — TODOs, dead code, naming issues. General-purpose tool for any project.
| name | save |
| description | Context preservation - route learnings to cal.md (permanent) or memories/ (ephemeral) |
| argument-hint | type content - Save type (delta, aha, memory, decision, session) and content to preserve |
Purpose: Preserve learnings and context with appropriate routing.
Arguments provided: $ARGUMENTS
Extract the learning, let the output vanish.
Agent outputs are scaffolding. Once you've acted on them, the output doesn't matter — only the extracted insight does.
| Type | Destination | Section in cal.md | Lifespan |
|---|---|---|---|
delta | cal/cal.md | ## Deltas | Permanent (Auto Dream prunes resolved) |
aha | cal/cal.md | ## Principles Learned | Permanent |
memory | cal/cal.md | ## Active Context | Prunable by Auto Dream when stale |
decision | cal/cal.md | ## Decisions | Permanent |
session | cal/memories/YYYY-MM-DD.md | — | Prunable |
# Permanent learnings → cal/cal.md
/cal:save delta "BELIEVED: X, ACTUAL: Y, DELTA: Z"
/cal:save aha "scope creep happens when..."
/cal:save memory "user prefers Socratic method"
/cal:save decision "chose X over Y because..."
# Ephemeral context → cal/memories/
/cal:save session # Full context dump for resume
Entries are atomic bullets appended to the correct section in cal/cal.md:
- **[topic] ([YYYY-MM-DD]):** BELIEVED: X. ACTUAL: Y. DELTA: Z.
- **[topic] ([YYYY-MM-DD]):** CHOICE: X. RATIONALE: Y. REVISIT-IF: Z.
- **[topic] ([YYYY-MM-DD]):** [principle or context]
One insight per bullet. Date in parentheses for Auto Dream pruning. No heading blocks.
Session saves go to cal/memories/YYYY-MM-DD.md:
## [TIME] SESSION — [Topic]
**Working on:** [Current task]
**Branch:** [Git branch]
**Uncommitted:** [Yes/No]
### This Session
- [What was accomplished]
### Resume With
- [Next steps]
### Hot Context
- [Critical details needed to continue]
| Output Type | Example | Do |
|---|---|---|
| Routine checks | "typescript: 0 errors" | Let vanish |
| Confirmations | "spec looks clean" | Let vanish |
| Agent reviews | 200-line review output | Extract learning, discard output |
On save, verify cal/cal.md exists. If not, create it with the full template including all four sections:
# Cal Brain
Cal's persistent project knowledge. Organized by topic, not chronologically.
Auto Dream consolidates entries between sessions — keep entries atomic and topical.
---
## Principles Learned
*Patterns promoted from repeated experience.*
---
## Deltas
*Wrong assumptions corrected. BELIEVED / ACTUAL / DELTA format.*
---
## Decisions
*Architectural choices with rationale. CHOICE / RATIONALE / REVISIT-IF format.*
---
## Active Context
*Current state that helps orient new sessions. Auto Dream prunes when stale.*
If cal/cal.md exceeds 300 lines, review and consolidate entries.
cal/cal.md is organized by topic (Principles, Deltas, Decisions, Active Context) not chronologically. This structure supports Auto Dream consolidation:
When saving, append to the correct section. Keep entries atomic — one insight per bullet.