Memory
Persistent memory: observations, session summaries, searchFrom its SKILL.md
npx -y skills add aAAaqwq/AGI-Super-Team --skill memoryAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
SKILL.md
5.4 KB, ~1.4k tokens by cl100k_base, as published. Nobody here has run it
Memory
Persistent memory system: automatic capture of observations, session summaries, memory search
When to use
- "what did I do yesterday?" / "what happened last week?"
- "remember this" / "save this decision"
- "memory stats" / "show memory"
- "search memory" / "find sessions about X"
Architecture
PostToolUse hook → observations (local JSONL)
SessionEnd hook → session summary (local + git)
SessionStart hook → inject context (auto)
Hybrid storage:
- Observations (raw) →
~/.claude-memory/observations/(local) - Session summaries →
$PROJECT_ROOT/memory/sessions/(git)
Paths
| What | Path |
|---|---|
| Observations (local) | ~/.claude-memory/observations/ |
| Sessions (local) | ~/.claude-memory/sessions/ |
| Sessions (git) | $PROJECT_ROOT/memory/sessions/ |
| Session index | $PROJECT_ROOT/memory/index.md |
| Search tool | ~/.claude-memory/hooks/memory-search.sh |
| Hook scripts | ~/.claude-memory/hooks/ |
| Config | ~/.claude-memory/config.json |
How to execute
Search observations
# By type
~/.claude-memory/hooks/memory-search.sh --type crm_update --days 7
# By project
~/.claude-memory/hooks/memory-search.sh --project $PROJECT_ROOT --days 30
# By tags
~/.claude-memory/hooks/memory-search.sh --tags crm,leads --days 14
# Text search
~/.claude-memory/hooks/memory-search.sh --query "Acme Corp" --days 30
# Full details
~/.claude-memory/hooks/memory-search.sh --query "Acme" --full
# By session
~/.claude-memory/hooks/memory-search.sh --session abc12345
Statistics
~/.claude-memory/hooks/memory-search.sh --stats
List sessions
~/.claude-memory/hooks/memory-search.sh --sessions
Save a decision manually
echo '{"id":"obs-'$(date +%Y%m%d)'-'$(openssl rand -hex 4)'","timestamp":"'$(date -u +%Y-%m-%dT%H:%M:%SZ)'","session_id":"CURRENT_SESSION","type":"decision","tool":"manual","summary":"DECISION_DESCRIPTION","context":{},"project":"$PROJECT_ROOT","tags":["TAG1"]}' >> ~/.claude-memory/observations/observations-$(date +%Y-%m-%d).jsonl
Read a specific session
cat $PROJECT_ROOT/memory/sessions/YYYY-MM-DD-SESSION_ID.md
Observation types
| Type | When |
|---|---|
feature | New functionality |
bugfix | Bug fix |
decision | Architectural/business decision |
discovery | New knowledge about the system |
change | General change (git, PR) |
outreach | Communication through channels |
crm_update | CRM data change |
config | Configuration change |
skill_update | Skill change |
retrospective | Quality assessment: what worked, what didn't, what to change |
Retrospective (convention)
At the end of a session or after a significant task, Claude records a retrospective observation:
echo '{"id":"obs-'$(date +%Y%m%d)'-'$(openssl rand -hex 4)'","timestamp":"'$(date -u +%Y-%m-%dT%H:%M:%SZ)'","session_id":"SESSION_ID","type":"retrospective","tool":"manual","summary":"RETROSPECTIVE TEXT","context":{},"project":"$PROJECT_ROOT","tags":["retro"]}' >> ~/.claude-memory/observations/observations-$(date +%Y-%m-%d).jsonl
What to record:
- What went wrong and why
- Which skill/tool didn't work or is incomplete
- What should be changed for the next session
- Important insights about the process
Summary format: one sentence, specific. For example:
"telegram-send: need to check bot access before bulk sending""change-review was skipped before PR — add a reminder""CSV parsing via csv.DictReader breaks on empty rows — use pandas"
When to record:
- Session had errors or rework
- Found a gap in a skill or tool
- Made a decision that affects future sessions
Retrospective automatically:
- Gets included in session summary (section
## Retrospective) - Gets injected into the next session (section
## Last retroin context)
Privacy
4 layers of protection:
<private>...</private>tags → stripped before saving- Pattern matching (token, password, api_key, Bearer)
- Sensitive files (.env, credentials.json, token.json) → ignored
- Sensitive commands (export TOKEN/SECRET) → redacted
Maintenance
Rotation (manual, once a month)
mkdir -p ~/.claude-memory/observations/archive
find ~/.claude-memory/observations -name "observations-*.jsonl" -mtime +30 -exec mv {} ~/.claude-memory/observations/archive/ \;
Disk usage
du -sh ~/.claude-memory/
du -sh $PROJECT_ROOT/memory/
Commit session summaries
cd $PROJECT_ROOT
git checkout -b update/memory-sessions-$(date +%Y-%m-%d)
git add memory/
git commit -m "Add session summaries for $(date +%Y-%m-%d)"
git push -u origin update/memory-sessions-$(date +%Y-%m-%d)
gh pr create --title "Memory: session summaries $(date +%Y-%m-%d)" --body "Auto-generated session summaries"
gh pr merge --squash --delete-branch
git checkout main && git pull
Hooks (automatic)
| Hook | Event | What it does |
|---|---|---|
session-start.sh | SessionStart | Inject context from past sessions |
post-tool-use.sh | PostToolUse (Edit/Write/Bash) | Records observations |
session-end.sh | SessionEnd + PreCompact | Generates session summary |
Related skills
daily-briefing— morning context (includes memory)show-today— today's tasksquery-leads— CRM searchlog-activity— activity logging (complements observations)
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.
Gives 0 of the 12 instructions most memory context skills give in ~1.4k tokens
Counted across 754 of the 1,056 authors here whose files we hold, read 2026-09-06
- Preserve existing content structurein 15 of 754, across 9 files
- Front-load the leading wordin 14 of 754, across 10 files
- Update existing entries instead of duplicatingin 14 of 754, across 7 files
- Keep CLAUDE.md under one hundred linesin 14 of 754, across 12 files
- Read CLAUDE.md at the project rootin 14 of 754
- Keep each meaning in a single source of truthin 12 of 754, across 8 files
- Redact sensitive information before committingin 11 of 754, across 4 files
- Scan for all CLAUDE.md filesin 11 of 754, across 7 files
- Use frontmatter for metadata on filesin 10 of 754, across 3 files
- Repeat user interactions 10 timesin 10 of 754, across 4 files
- Write the CLAUDE.md file into the target folderin 10 of 754, across 8 files
- Use memlab to process snapshotsin 9 of 754, across 3 files
Said here and by no other author read
- Save decisions manually to observations
- Record retrospective observation after significant tasks
Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.