← The Cosmic Store
C
Continuity Kit Pro v1.0.0 · macOS
15 independent reviews 108 automated tests · zero open bugs

Stop losing context
every Claude Code session.

A bash + markdown kit that gives Claude Code guaranteed session-start context, cross-machine memory through git, and guard rails against the dev commands that crash a 16 GB Mac when multiple sessions are open.

5-min setup · 30-day refund · free 1.x updates · single-developer license

Built and used daily by Pedro across his own 40+ project ecosystem.

~/code/your-project $ claude

# SessionStart hook fires automatically:
╔══════════════════════════════════════════════════╗
║  CANONICAL CONTEXT AUTO-INJECTED              ║
║  Read this BEFORE your first tool call.        ║
╚══════════════════════════════════════════════════╝
▼ CLAUDE.md ........................ 1.2 KB
▼ docs/SESSION-PROTOCOL.md ......... 3.8 KB
▼ Last SESSION-LOG (2026-04-25) .... 2.1 KB

# Same machine. Same project. Same session. Hot.

~/code/your-project $ npx next dev
🚫 BLOCKED — pattern flagged in .claude-memory-kit.guard
Why: Next.js dev under Turbopack uses 3-6 GB RAM.
With multiple Claude sessions open, your Mac will crash.

~/code/your-project $ ./scripts/sync-claude-memory.sh to-repo
→ memory committed to git, ready for the other machine.

What every session looks like once the kit is installed.

The 10–20 minute tax you pay every session

Claude Code starts every session cold. If you work across multiple projects or machines, it's worse:

Session A closes with context you'll need later → lost the moment the window closes.

Machine B has memories Machine A doesn't → out of sync, no warning.

The model reads CLAUDE.md? Maybe. It's an instruction, not a guarantee.

Commits pile up with no narrative → next session reads git log and guesses.

What's inside the ZIP

Drops into any project on macOS in 5 minutes. No dependencies beyond what macOS already ships.

SessionStart injector

Hook-level. Cats your canonical docs (CLAUDE.md, SESSION-PROTOCOL, latest SESSION-LOG) into every session. Not "best effort" — guaranteed.

Memory-hungry guard

Blocks next dev, npm run dev, metro, vite, expo start, turbo dev — including wrappers (bash -c, sudo, ;, &&) and path prefixes (./node_modules/.bin/...). Per-command bypass via env var.

Cross-machine sync

Memory mirrors to docs/CLAUDE-MEMORY/ in your repo, travels with git, lands on Machine B. Conflict detection when both sides edit. Pre-mutation backups to a durable path.

Session-end audit

Warns you on close if the SESSION-LOG is missing, memory diverged, commits are unsynced, or files are pending. Writes a SESSION-CHECK report the next session reads.

SESSION-LOG generator

Builds a skeleton from git log so the next session has a real handoff, not an archaeology dig. Atomic file claim — safe under parallel invocations.

Post-install verifier

26 automated checks the buyer can run on demand. Plus 25 end-to-end smoke tests and a 57-assertion regression matrix.

Also shipped:

13 memory templates with {{PLACEHOLDERS}} · 6 docs (SESSION-PROTOCOL, TROUBLESHOOTING, CUSTOMIZING-HOOKS, MIGRATION-FROM-V1, UNINSTALL, PRIVACY-AND-SHARING) · LICENSE, SECURITY, CHANGELOG · CHECKSUMS.txt with SHA-256 of every shipped file

But Claude Code already has /memory — why pay €120?

Built-in /memory covers ~40 % of what serious Claude Code users need. This kit covers the other 60 % that matters when you use the CLI across multiple projects and machines.

Capability Built-in /memory This kit
Same machine, same project
Cross-machine sync (Mac ↔ WSL ↔ Linux, via git)
Guaranteed context injection at session start❌ best effort✅ hook-level
Blocks memory-hungry commands that crash your Mac
Audits session close with carry-forward notes
Conflict detection when two machines edit the same memory
Auto SESSION-LOG skeleton from git log
Pre-mutation backups to a durable path

This is not a weekend hack.

15 independent code reviews across macOS bash 3.2 and WSL Ubuntu bash 5.2. 78 issues surfaced — 71 fixed in 1.0.0, 7 deferred to [Unreleased] as documented design decisions or planned 1.1 additions. None are open bugs in normal use. The full hunting log lives in CHANGELOG.md.

set -euo pipefail killing the SessionStart hook when local memory diverged from the repo (the exact case the hook exists for). Fixed in round 6.

6 ways to wrap next dev to slip past a naive guard. All blocked via a segment-aware Python parser. Fixed in round 6.

./node_modules/.bin/expo start sliding through a path-prefix guard — the way npx/pnpm exec resolve in real projects. Fixed in round 11.

A to-repo --force shape that wiped the repository when local was empty. Now refused by default, requires explicit env-var opt-in. Fixed in round 11.

A cp -Rp missing a single space silently no-op'ing the safety backup. Fixed in round 13.

CRLF line endings in the config breaking every hook with cryptic exit-127. tr -d '\r' everywhere. Fixed in round 10.

Round 15 was an independent privacy audit. Zero bugs, zero network calls, zero credentials surfaced. Its checklist shipped as docs/PRIVACY-AND-SHARING.md so you can run the same review yourself before pushing the project to a public repo.

🐾🌿

Why this product exists

Every hour I save by not rebuilding context in the terminal is an hour I can give to the animals.

€80
per sale
to the animal sanctuary fund 🐾
€20
per sale
sustains me, infra, payment fees

The €120 listing price covers Gumroad and Stripe fees, so what reaches me splits cleanly: €80 to the animals, €20 to keep this going. No discounts, no fake urgency. The kit is sold to make the sanctuary possible — not to chase volume.

FAQ

Will it run on Linux or WSL2?

Yes — the scripts are POSIX bash and we cross-validate on WSL Ubuntu bash 5.2. But macOS is the officially supported platform; Linux/WSL is best-effort. Native Windows (cmd / PowerShell) does not work — Windows users need WSL2.

Does it phone home? Telemetry? Auto-update?

No. There is no curl, no wget, no analytics endpoint, no auto-update. grep -rE 'curl|wget|nc |https?://' scripts/ matches only documentation references. Data leaves your machine only when you run git push yourself.

What's the security model?

One footgun: .claude-memory-kit.config is sourced as bash by every hook. Anything written there runs on your machine when a Claude session starts. In a shared repo, review config changes on every pull — same care you'd give a Makefile. For solo use it's a non-issue. Full discussion in SECURITY.md.

Can I uninstall cleanly?

Yes — docs/UNINSTALL.md ships with a step-by-step. Path A removes everything; Path B keeps your accumulated memory but unhooks the kit. No system-wide changes, no ~/.bashrc/~/.zshrc edits to undo.

What's NOT included?

No GUI (shell + markdown by design), no semantic search across memory (planned for 2.x — 1.0 is grep-friendly markdown on purpose), no SaaS. No 1:1 onboarding at this price tier — email questions get an answer; deep customisation is yours.

License?

Single-developer commercial license. Use on your own and your employer's projects freely. Don't resell the kit itself; redistribution of the patterns (writing your own similar kit) is fine — the value is the engineering, not the idea. Full text in LICENSE.md.

Refund?

30 days, no questions. Email support@thecosmicapp.com with your order ID and you get a full refund within 5 business days.

Ready to keep your context?

5 minutes to install. Every Claude Code session, hot from then on.

Buy on Gumroad — €120

30-day refund · free 1.x updates · single-developer license · 78108-byte ZIP, SHA-256 2acae91e…