remember
npx machina-cli add skill shane9coy/katana-agent/remember --openclawRemember — Store Memory
When This Fires
- User types
/remember - User says "remember this", "save that", "note that down"
- Auto mode: session is ending
What To Do
Step 1: Classify the memory
| Conversation contains... | Write to... |
|---|---|
| How the agent should behave | ~/.katana/memory/core/soul.md → Learned Behaviors |
| User preferences, goals, personal facts | ~/.katana/memory/core/user.md → appropriate section |
| Work done, code changes, decisions | ~/.katana/memory/work.md → new entry at top |
If session contains multiple types, write to multiple files.
Step 2: Summarize
- Strip raw code blocks longer than 5 lines
- Strip tool call outputs and error logs
- Strip repetitive troubleshooting
- Keep: decisions, problems solved, features built, user preferences
- Condense to 1-3 sentences for work.md
Step 3: Detect project context
- If user specified:
/remember --project myapp→ use that - If inside a git repo: use repo folder name
- If neither: ask user or tag as
general
Step 4: Write
work.md — prepend at TOP:
## YYYY-MM-DD — project-tag
Summary sentence(s).
user.md — find right section, append:
- YYYY-MM-DD: New context or preference noted here.
soul.md — append to Learned Behaviors:
- NEW: Description of learned behavior
Step 5: Confirm (manual mode)
Before writing, show the user:
📝 Memory save:
→ work.md: "Summary of what happened"
Tagged: project-name | YYYY-MM-DD
Save this? (y/n/edit)
Step 6: Acknowledge
✓ Saved to work.md (project-name, YYYY-MM-DD)
Source
git clone https://github.com/shane9coy/katana-agent/blob/main/templates/kilocode/skills/remember/SKILL.mdView on GitHub Overview
Remember captures a memory from the current session and saves it to the appropriate memory file. It triggers on /remember, 'remember this', 'save that', or auto mode at session end. The tool classifies content into soul, user, or work memories, summarizes key details, detects the project, and writes to the correct memory file.
How This Skill Works
It follows a five-step workflow: classify memory to soul.md, user.md, or work.md; summarize and strip long blocks or logs; detect project context via a flag, a git repository, or general tag; write to the appropriate file (work.md at the top, others appended); and optionally prompt for confirmation in manual mode.
When to Use It
- Ending a session and wanting to preserve what happened
- Capturing a user preference, goal, or personal fact
- Recording work details like decisions, code changes, or features built
- Remembering a specific instruction or behavior the agent should learn
- Auto mode when the session ends and memory needs persistence
Quick Start
- Step 1: Trigger memory capture with /remember or say 'remember this'.
- Step 2: Let the agent classify, summarize, and detect the project context.
- Step 3: Review the auto/manual prompt and save to work.md, soul.md, or user.md.
Best Practices
- Ensure each memory is classified to the correct file (soul, user, work)
- Keep work.md summaries to 1-3 sentences; strip code blocks and logs
- Tag with project context (explicit --project or inferred repo name)
- Avoid storing sensitive data; anonymize where possible
- In manual mode, review prompts before saving; confirm changes
Example Use Cases
- 2026-03-08: New preference saved: user wants concise summaries.
- 2026-03-08: Work recorded: implemented memory logging for tasks.
- 2026-03-07: Learned behavior: remember mode should respect privacy.
- 2026-03-08: Project detected: myapp; memory written to work.md top.
- 2026-03-08: User instruction noted: remember this to reinforce behavior.
