Engineering guardrails that keep AI coding agents on track.
The plugin follows the Agent Plugins standard where it can, but the
standard is not yet adopted by other vendors. Rules and agents live in com.github.copilot/,
the only directory VS Code reads them from, so the plugin is VS Code first. Skills are plain
SKILL.md files and travel to other agents more easily than rules do.
To install railing plugin into VS Code use this link:
vscode://chat-plugin/install?source=SavenkovIgor/railing&plugin=railing
It will install both, plugin and marketplace (for autoupdate) with this plugin only
├── plugin.json plugin name, version, description, Agent Plugins schema
├── marketplace.json marketplace listing that lets VS Code install the plugin
├── com.github.copilot/ the only directory where VS Code reads rules and agents
│ ├── agents/
│ │ ├── controller.agent.md owns the task, runs planner, coder, and reviewer
│ │ ├── planner.agent.md turns a brief into a plan
│ │ ├── coder.agent.md implements a plan
│ │ ├── reviewer.agent.md checks a change against its task
│ │ ├── researcher.agent.md experimental, researches a question and reports options
│ │ └── gilfoyle.agent.md blunt code review and analysis
│ └── rules/
│ ├── global.instructions.md general agent behavior and workflow guidance
│ ├── issues.instructions.md scope of one issue in any tracker
│ ├── code.instructions.md shared coding principles for C++, Python, JavaScript, TypeScript
│ ├── code-cpp.instructions.md C++ coding style
│ ├── code-python.instructions.md Python coding style
│ ├── markdown.instructions.md formatting checks for all *.md files
│ ├── prose.instructions.md how developer-facing prose reads, in all *.md files
│ ├── notes.instructions.md path, task ID, and language of agent-written notes in notes/
│ ├── reports.instructions.md content requirements for reports about existing code in notes/
│ └── reports-overview.instructions.md structure and checks for *.overview.md reports
└── skills/
├── bp/ validate files against best-practice references
├── debug-context-artifacts/ inspect the initial AI context state
├── decision-flow/ walk through open decisions in short rounds of questions
├── design-system/ create or audit design systems
├── engineering-maturity/ rank missing engineering practices by return on effort
├── migration-skill-factory/ create skills for codebase migrations
├── project-glossary/ merge tracker, docs, and code terms into GLOSSARY.md
├── reflect/ propose context improvements after a task
├── report-review/ clean up a report and verify it against the code
├── report-write/ write a technical report about existing code into notes/
├── review/ review a change against its intent, report only
├── to-issues/ convert plans and discussions into tracker issues
└── tour/ explain a codebase: one-page map, then a drill into one area
Which rule and skill files depend on which. A solid arrow means "links to". A dashed arrow means that the applyTo scope of the file lies inside the scope of the target, so both attach together.
flowchart TD
subgraph SK["Skills"]
BP["skill: bp"]
RR["skill: report-review"]
RW["skill: report-write"]
RV["skill: review"]
TR["skill: tour"]
end
subgraph IN["Instructions"]
NT["instruction: notes"]
REP["instruction: reports"]
RO["instruction: reports-overview"]
PR["instruction: prose"]
MD["instruction: markdown"]
end
BP --> RR
RV --> BP
BP --> PR
BP --> MD
RR --> NT
RR --> REP
RW --> RO
TR --> REP
RO -.-> REP
REP -.-> NT
NT -.-> PR
NT -.-> MD