More screenshots
Works with the AI Agents you already use
Claude Code Memory Saver: Rules Survive Compaction
Turn /compact from a reset into a checkpoint: your rules, goal, edited files and last test result are saved before compaction and put back after it.
$12
About this skill
After auto-compact the agent forgot the rule you gave an hour ago, redid a finished step, and called a test run from before its last edit "green". Measured: on django (7,089 files) the checkpoint is built in 0.7 s and fits in 1,019 bytes. In real headless Claude Code sessions every hook fired and the checkpoint came back right after /compact.
Context Checkpoint keeps your working state outside the model. Hooks record it as you work, save it before compaction, and put it back into context after:
- Your rules, verbatim. "Never touch migrations/", "use pnpm, not npm" and "don't commit" are captured from your prompts by deterministic patterns (English and Spanish), or from anything written as
RULE:. Project rules can be pinned for every session. - A check on the summary. After compaction it compares your rules with Claude Code's own summary and marks any rule the summary left out.
- The truth about tests. The last test command, its exit code and summary line, flagged STALE when code changed afterwards, or when a pipe hid the exit code.
- What's done. The todo plan (now / next / done, do not redo), unresolved errors, uncommitted files with +/- lines, and the commits made this session.
- A hard 1.5 KB budget, filled by priority.
show,diffandsavegive the same view on demand, for handoffs and resumes.
It never parses the internal transcript, never calls a model, and never blocks compaction. The recording hooks run async. One command installs the hooks, merging into your settings without overwriting anything.
| Benchmark | click (178 files) | django (7,089 files) |
|---|---|---|
| Changed files named | 7/7 | 4/4 |
| Checkpoint size | 1,299 B | 1,019 B |
| Edit hook, median | 127 ms | 179 ms |
| Save before compaction | 0.79 s | 0.70 s |
Rule capture on 80 prompts written by a separate agent that never saw the code: 41 of 52 rules on the first, untuned pass, then 50 of 52 with 1 false capture after tuning.
Honest result: in our two A/B runs (real Claude Code, Haiku, 5 and 13 prompts), Claude Code's own summary also kept every rule, so both variants scored 8/8. What you buy is a guarantee that does not depend on the summary, plus the test, plan and error state it does not track.
It does not replace the summary, read your transcript, or catch rules phrased in unusual ways (use RULE: for those).
Works in Claude Code, Cursor, Codex CLI, Copilot and any SKILL.md agent. The hooks need Claude Code; other agents use show and save.
Privacy: stdlib Python, no network access, no telemetry. State stays in your project's git-ignored .claude/checkpoint/ folder.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse All Claude Skills.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 27 days ago
- Software
Python
Git