Turn every coding session into a structured learning trail.
A collection of Agent Skills that generate structured learning entries as a byproduct of AI-assisted coding sessions. As the agent writes code, the skills simultaneously document the reasoning, patterns, and language features used -- producing a cumulative, browsable knowledge base in a .decodie/ directory.
Compatible with 70+ AI coding agents including Claude Code, Gemini CLI, Cursor, Cline, Windsurf, and more.
While you code with an AI agent, the Decodie skills observe each meaningful decision the agent makes and write structured learning entries capturing:
- What the code does (code snippets, key concepts)
- Why this approach was chosen (rationale, alternatives considered)
- Where it lives (content-based code references that survive refactoring)
- Related resources (links to official docs for PHP, JavaScript, Python, React, and more)
Entries are tagged by experience level (foundational through advanced), decision type (explanation, rationale, pattern, warning, convention, overview), and topic. Duplicate concepts are detected and cross-referenced automatically.
Install with the Skills CLI:
# Install all skills
npx skills add owenbush/decodie-skill --all
# Install a specific skill
npx skills add owenbush/decodie-skill -s decodie-analyze
# Install for a specific agent
npx skills add owenbush/decodie-skill --all -a claude-codenpx skills add owenbush/decodie-skill --all -gUsing the deprecated install-skill command
The legacy installer still works but will be removed in a future release:
npx @owenbush/decodie-ui install-skill
npx @owenbush/decodie-ui install-skill --scope projectThe skills create a .decodie/ directory at the project root with the following structure:
.decodie/
├── config.json # User preferences (experience level, topic filters)
├── index.json # Lightweight index of all entries (metadata only)
└── sessions/
├── 2026-03-27-001.json # Full entries from session 1
├── 2026-03-27-002.json # Full entries from session 2
└── ...
index.json-- contains metadata for every entry: title, topics, experience level, code references, external doc links, and lifecycle state.sessions/*.json-- contain the full content of each entry: code snippets, explanations, alternatives considered, and key concepts.config.json-- user preferences such as preferred/excluded topics and archival thresholds.
You can commit .decodie/ to share learning entries with your team, or add it to .gitignore to keep it personal.
| Skill | Description |
|---|---|
decodie-observe |
Document decisions as you code in real-time |
decodie-analyze |
Analyze existing code retroactively (selective or exhaustive) |
decodie-overview |
Generate a high-level summary of a file, directory, or project |
decodie-explain |
Explain a code selection (ephemeral by default) |
decodie-ask |
Ask questions about existing entries |
decodie-verify |
Confirm entries still match source code and stamp with commit SHA |
decodie-flag-stale |
Fast CI-friendly check for entries affected by recent changes |
Install the Decodie VSCode extension to browse entries in your editor sidebar, with gutter indicators, right-click analysis, and entry detail views.
npx @owenbush/decodie-ui serveOpens a browsable interface at http://localhost:8081 with lessons, progress tracking, and Q&A.
ddev add-on get owenbush/decodie-ddev
ddev restart && ddev decodieJSON schemas for all data files are in schema/:
schema/index.schema.json--.decodie/index.jsonschema/session.schema.json--.decodie/sessions/*.jsonschema/config.schema.json--.decodie/config.json
See schema/README.md for detailed documentation.
- decodie.owenbush.dev -- Project homepage
- owenbush/decodie-ui -- Web-based browser with lessons and progress tracking
- owenbush/decodie-vscode -- VSCode extension
- owenbush/decodie-github-action -- GitHub Action for automatic PR analysis (Marketplace)
- owenbush/decodie-github-bot -- Interactive bot for PR comments
- owenbush/decodie-ddev -- DDEV add-on
- owenbush/decodie-core -- Shared data layer (types, parser, reference resolver)