From memory-keeper
This skill should be used when the user says "context save", "remember this", "save this", "note this for later", "keep this", or wants to persist any knowledge, pattern, decision, or learning from the current session to persistent memory.
How this skill is triggered — by the user, by Claude, or both
Slash command
/memory-keeper:context-saveThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
`/context save <what to remember>`
/context save <what to remember>
Examples:
/context save ContextDomain is a new field for matching block outputs by domain context/context save guard findSessionJsonl against undefined sessionId — was matching real files/context save (will ask what to save)Read insights_root from ~/.claude/memory-keeper.local.md YAML frontmatter. If the file is missing, stop and ask the user to create it with the required settings (see plugin README).
Each entry gets one classification that determines where it's saved:
| Classification | What it is | Saved to |
|---|---|---|
insight | Completed work — code changes, patterns applied, decisions made, gotchas | <insights_root>/<repo>/insights.md |
agent_edit | Changes to AI behavior — agent guards, skill descriptions, hook logic, prompts, CLAUDE.md, plugin config | <insights_root>/claude-config/behavior.md |
task | ONLY unstarted intentions — "I need to refactor X", "TODO: add Y" | <insights_root>/_tasks/pending.md |
One save invocation may produce multiple entries with different classifications.
Each entry is a topic heading + bullet-pointed facts. Each fact = WHAT changed + WHY/FOR WHAT.
## <topic> — <YYYY-MM-DD HH:MM>
- <fact 1: what was done + why it matters>
- <fact 2: what was done + why it matters>
Good facts (concise, high-level, with WHY):
Bad facts (too granular or vague):
Derive <repo> by running git -C <cwd> rev-parse --show-toplevel 2>/dev/null and taking the basename. If not a git repo, use the basename of cwd. The user or context may override (e.g. if the conversation clearly refers to a different project).
Before saving an insight, check <insights_root>/_tasks/pending.md for an active task (status: active).
<insights_root>/_tasks/<task-slug>/notes.md instead of the project folder.<insights_root>/<repo>/insights.md as usual.agent_edit and task entries always go to their fixed locations regardless of active task.
Before appending any entry, read the target file and check for duplicates:
## <topic>) already existsOne precise entry is better than two fuzzy ones.
<insights_root>/<repo>/insights.md<insights_root>/claude-config/behavior.md<insights_root>/_tasks/pending.md<insights_root>/_tasks/<task-slug>/notes.mdnpx claudepluginhub popoffvg/claude-plugin-memory-keeper --plugin memory-keeperPersists learnings into a 5-layer memory hierarchy (CLAUDE.md files, memory/MEMORY.md) and consolidates by pruning outdated entries and promoting recurring patterns. Triggers on 'extract learnings', 'remember', 'dream'.
Saves key project knowledge explicitly to MEMORY.md via /si:remember for reliable recall. Checks duplicates, warns on size, suggests CLAUDE.md promotion.