FAQ and troubleshooting
- About RDF
- What is RDF?
- Who is it for?
- How is this different from a CLAUDE.md?
- How does it compare with Spec Kit, BMAD or Superpowers?
- Which AI tools does it support?
- Is it worth it on a small project?
- Will it bloat my context?
- Does RDF send my code anywhere?
- What gets committed to my repository?
- What is the difference between
rdf initand/r-init?
- Troubleshooting
- Update and uninstall
- Still stuck?
About RDF
What is RDF?
RDF (rfxn Development Framework) is a governance layer for AI coding agents. It keeps your project’s conventions in files the agent reads, adds a spec-to-ship pipeline, and has separate agents review the work. It is unrelated to the W3C Resource Description Framework.
Who is it for?
Anyone whose code matters enough to review, whether that is a solo maintainer or a team shipping to production. It came from maintaining security tooling, but profiles cover Python, Go, Rust, TypeScript, Node, PHP, Perl, frontend, databases and infrastructure.
How is this different from a CLAUDE.md?
A CLAUDE.md is one file you write and maintain. RDF generates one from your stack’s profiles, then adds what a single file cannot do: governance derived from your actual code, a reviewer that challenges plans before implementation, gates that demand evidence, and a doctor that detects drift.
How does it compare with Spec Kit, BMAD or Superpowers?
They share the spec-first idea. RDF’s focus is governance generated per repository, adversarial gates sized to each change’s scope and risk, a CLI for project health and token spend, and one source deployed to several AI tools. If you want only a spec template and a task list, a lighter tool may suit you better.
Which AI tools does it support?
Claude Code, Codex and Antigravity CLI are first-class. Any tool that reads AGENTS.md gets the documentation surface, and Gemini CLI is a frozen legacy tier. Claude Code has all 38 commands, while Codex and Antigravity get the 11 lifecycle commands. See Multi-tool parity.
Is it worth it on a small project?
Often, yes, because the ceremony scales. /r-plan --bugfix writes a short failing-test-first plan and keeps gates minimal, and you can skip the pipeline and run /r-verify or /r-review on any diff. See Pick a tier.
Will it bloat my context?
The default deploy adds ~0.1K always-loaded tokens per session. Your scaffolded CLAUDE.md costs more (about 3K tokens on the Quickstart’s Flask app), and skill and agent descriptions add about 1.6K by the same estimate. Governance files load on demand. See Token usage.
Does RDF send my code anywhere?
RDF has no telemetry and no RDF-hosted accounts, and it makes no network calls to R-fx Networks. Your code goes only to the AI tool you already use, under that tool’s terms. See the privacy policy for the version and pricing lookups /r-env can make.
What gets committed to my repository?
rdf init adds CLAUDE.md, PLAN*.md, AUDIT.md, MEMORY.md, .rdf/ and .agents/ to .git/info/exclude, so they stay local. Specs in docs/specs/ and plans in docs/plans/ are normal files meant to be committed, and SECURITY.md and CONTRIBUTING.md are yours to keep or delete.
What is the difference between rdf init and /r-init?
rdf init runs in your shell with no model calls. It writes CLAUDE.md, reference docs and the .rdf/ layout. /r-init runs in your agent session, scans your code and writes the project governance. Use both, in that order. Plugin installs have no CLI, so they use /rdf:r-init alone.
Troubleshooting
rdf: command not found
The checkout is not on your PATH. Link it:
mkdir -p ~/.local/bin && ln -s ~/rdf/bin/rdf ~/.local/bin/rdf
Make sure ~/.local/bin is on your PATH, or use bin/rdf from the checkout.
manual merge required: hooks.json
This is expected on a symlink deploy and does not affect the exit status. RDF does not edit your settings.json. Follow the merge step to turn on hooks and the status line.
/r-start shows no governance
Run /r-init in the project. rdf init only scaffolds; /r-init writes the governance files.
/r-init says governance already exists
Run /r-refresh to update it, which preserves your edits. /r-init --force deletes .rdf/governance/ and regenerates from scratch, which also removes the reference docs rdf init copied there.
A plugin command reports missing helpers
A SessionStart hook installs the ~/.rdf/state/ helpers. Restart Claude Code once after installing or updating the plugin.
The status line is blank
Hooks and the status line need jq on your PATH, and the symlink deploy needs the hooks merge. See the status line reference.
rdf doctor shows WARNs
WARNs are advisory, and a fresh install usually has a few. Only FAIL findings set a non-zero exit. Run rdf doctor --quiet to see just WARN and FAIL rows.
rdf doctor --all refuses to run
Your checkout sits directly in $HOME, and --all will not scan a home directory. Point it at your projects with RDF_WORKSPACE=~/projects rdf doctor --all.
Update and uninstall
How do I update?
From source:
cd ~/rdf && git pull && bin/rdf generate claude-code && bin/rdf deploy claude-code
Plugin: update it from /plugin, then restart Claude Code.
How do I uninstall?
Plugin:
/plugin uninstall rdf@rdf
From source, remove the symlinks and RDF’s state:
rm ~/.claude/skills/r-* ~/.claude/skills/reference
rm ~/.claude/agents ~/.claude/scripts ~/.claude/reference ~/.claude/governance
rm -rf ~/.rdf
If you merged hooks, remove the hooks entries and statusLine that point at ~/.claude/scripts/ from ~/.claude/settings.json.
From a project, delete .rdf/, CLAUDE.md and MEMORY.md, then remove the RDF working files block from .git/info/exclude.
Still stuck?
Search discussions or open an issue. Support is best-effort and community-driven.