Keep Learning#

Learn knowledge from local directories and store it in agent memory for future reference.

When to Use This Skill#

Activate this skill when user says:

• "持续学习知识"
• "keep learning"
• "learn knowledge base"
• "学习知识库"

Supported File Formats (v0.0.1)#

FormatExtensionsSupportMarkdown.md, .markdownFullPython.pyFullJavaScript/TypeScript.js, .ts, .jsx, .tsxFullJava.javaFullGo.goFullRust.rsFullC/C++.c, .cpp, .h, .hppFullShell.sh, .bash, .zshFullYAML/JSON/TOML.yaml, .yml, .json, .tomlFullSQL.sqlFullOther text.txt, .csvFull

Not supported in v0.0.1: PDF, Word, Excel, PowerPoint, Keynote, audio, video files.

Three-Layer Knowledge Architecture#

LayerStorageContentPurposeL1 Core MemoryAgent MemoryKey conclusions, core concepts, decisionsAuto-surface in daily conversationsL2 Knowledge IndexAgent MemoryFile paths, summaries, keyword mappingsKnow where knowledge livesL3 Source FilesLocal filesystemComplete original contentDeep-dive when needed via read_file

How It Works:

1. Daily conversations: L1 memories automatically appear in memory_overview
2. Need more detail: Query L2 index to find relevant files
3. Deep investigation: Use read_file to access L3 source files

Runtime Data Directory#

All runtime data is stored in ~/.keep-learning/:

FilePurposelast-commitGit commit hash of last learning sessionconfig.jsonUser configuration (knowledge base path, etc.)

Learning Workflow#

Step 1: Get Configuration#

First, search memory (category: project_environment_configuration) for an existing knowledge base path.

• If found: confirm the path with the user before proceeding. Example: "Found your knowledge base at ~/knowledge/work-assistant. Start learning from there?"
• If NOT found: stop and ask the user to provide the knowledge base path before doing anything else. Do NOT proceed until the user provides a valid path. Example: "Please provide the path to your knowledge base directory (e.g., ~/knowledge/work-assistant)."

Once confirmed, store the path in memory using update_memory with category project_environment_configuration.

Step 2: Git Pull (If Applicable)#

Check if knowledge base is a git repository and pull latest changes before learning.

Step 3: Scan Files#

Scan for supported files. Exclude: .git, node_modules, .obsidian, pycache, .venv

Step 4: Detect Changes (Incremental Learning)#

For git repositories, detect ALL types of changes:

1. Committed changes: Compare current HEAD with last-commit hash stored in ~/.keep-learning/last-commit using git diff <last-commit> HEAD --name-only
2. Uncommitted changes: Detect modified/added files in working directory using git status --porcelain

Combine both results to get the full list of changed files. This ensures learning happens even when:

• Remote has no updates, but local files were edited
• Local commits exist that haven't been pushed yet
• Files are modified but not yet committed

After learning completes, update ~/.keep-learning/last-commit with current HEAD hash.

For non-git directories: scan all supported files (no incremental detection).

Step 5: Read and Extract Knowledge#

For each file: read content, identify theme/concepts/conclusions, extract key knowledge.

Step 6: Store L1 Core Memory#

Create L1 memory entries using update_memory with appropriate category:

• expert_experience: Domain expertise, best practices
• project_introduction: Project/product overviews
• learned_skill_experience: Reusable methods, procedures

Title format: [Domain] Concise Topic Description

Step 7: Build L2 Knowledge Index#

Create knowledge index with file path, theme, keywords mappings.

Step 8: Generate Learning Report#

Output: Timestamp, Statistics, L1 Memories list, L2 Index summary, Notes.

Memory Deduplication#

Before creating: search_memory first. If exists, update; if not, create.

Quick Reference#

SituationActionFirst time userAsk for knowledge base pathGit repo detectedRun git pull before scanningLarge fileRead in chunks, summarize each sectionDuplicate knowledgeUpdate existing memoryUnsupported fileSkip and note in report

Limitations (v0.0.1)#

• Only Markdown and code files supported
• No PDF/Word/Excel/PPT support
• Memory entries have size limits