Recall past work from Claude Code history before starting new tasks.
Use when: (1) starting work on a codebase, (2) asked about previous decisions,
(3) need context about why code exists.
Installs into .claude/skills of the current project.
Are you the author of Engram Recall?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/majiayu000-engram-recall-claude-skill-registry)
---
name: engram-recall
description: |
Recall past work from Claude Code history before starting new tasks.
Use when: (1) starting work on a codebase, (2) asked about previous decisions,
(3) need context about why code exists.
category: memory
disable-model-invocation: false
user-invocable: true
allowed-tools: Read, Bash, Glob
---
# Engram Recall
Before diving into implementation, recall relevant context from past sessions.
## When to Use
- Starting work on an existing codebase
- Asked "why was this done this way?"
- Need to understand previous decisions
- Continuing work from a previous session
## Workflow
### Step 1: Check for engram database
```bash
ls .engram/memory.db 2>/dev/null || echo "No engram database found"
```
If no database exists, ingest Claude Code history:
```bash
npx engram ingest-claude --days 14
```
### Step 2: Search for relevant context
Before implementing, search for related decisions:
```bash
npx engram search "your task keywords" --json
```
### Step 3: Review file-specific context
If working on a specific file:
```bash
npx engram context get --file path/to/file.ts --json
```
### Step 4: Document new decisions
After making significant decisions, add context:
```bash
npx engram context add --file path/to/file.ts --note "Chose X over Y because..."
```
## Integration with Other Skills
This skill chains well with:
| Skill | When to Combine |
|-------|-----------------|
| **tdd** | Recall past test patterns before writing tests |
| **code-review** | Check if similar code was reviewed before |
| **describe-codebase** | Augment description with decision history |
## Output Format
When recalling context, format findings as:
```
## Relevant Context Found
**From session [date]:**
- [Summary of decision or discussion]
- Related files: [list]
**Recommendation:** [How this affects current task]
```
## Important
- Don't blindly follow past decisions if they were wrong
- Context is for understanding, not for copy-paste
- Update context when decisions change