Open locally
The CLI starts on a random localhost port with a per-session bearer token.
Local control center for AI agent configuration
Your AI agents have config. Now it has a control center.
Run one command in any repo to inspect every detected runtime, find real configuration problems, and apply fixes through a diff you review. Nothing leaves your machine.
$npx agentconfiging
Runs on your machine · no account · nothing uploaded · MIT
Proof, not promise
See the finding. See the fix. Then decide.
ERRORLOCAL
Selected finding · settings-local-committed
.claude/settings.local.json may be committed
This local settings file is present, but the project has no .gitignore. Local settings often contain personal tokens or opt-ins that shouldn’t be shared.
Proposed change
.gitignore
+.claude/settings.local.jsonExact file diff ready for review
3 agents detectedfixture-backed demonstrationlocalhost only
Reproducible from the sanitized project in site/fixtures/demo-repo. The window combines real analyzer output into one illustrative review state; the disabled action does not modify this page.
How it works
Run. Inspect. Fix.
See the whole repo
Detected runtimes, parsed artifacts, MCP servers, instructions, hooks, and findings in one UI.
Review every write
Machine-fixable findings produce an explicit diff before anything changes on disk.
Feature map
The same mental model from first visit to daily use.
Find what matters
Browse every artifact and run analyzers for broken imports, stale rules, risky settings, and more.
Change with context
Edit instructions, hooks, MCP servers, skills, rules, and settings with a dry-run diff.
Keep one source
Generate runtime instruction formats from a chosen source and turn drift into a visible finding.
Install traceably
Use an offline-first catalog with checksums, provenance, and reviewable changes.
Read local history
Understand activity and context health from redacted histories already on your machine.
Stay in the loop
Work through git, terminals, and visual pipelines without losing sight of the underlying files.
Runtime coverage
Eight detected. Seven more instruction targets.
Claude Code
Cursor
GitHub Copilot
Codex
Continue
Aider
Gemini CLI
opencode
Cline
Windsurf
Zed
Amazon Q
Junie
Roo
Qodo
Agent changes over time
Upstream behavior moves. The history stays visible.
Content-safe before-and-after summaries from official profile refreshes. Observed changes stay pending until human review promotes them.
Override instruction files observed
Codex documentation now identifies explicit global and project override files.
- Before
- Instructions were represented by global and project AGENTS.md files.
- After
- The profile also recognizes ~/.codex/AGENTS.override.md and project AGENTS.override.md.
Project rules directory observed
Claude Code documentation now identifies a project rules directory alongside CLAUDE.md.
- Before
- Project instructions were represented by CLAUDE.md files.
- After
- Projects can also organize scoped Markdown rules under .claude/rules/*.md.
Security model
Private by architecture, not by policy.
Nothing leaves your machine.
No hosted backend. No telemetry. No account. Secrets are redacted server-side before config content reaches the browser.
- 127.0.0.1 on a random port
- Per-session bearer token
- Canonical-path write guards
- Dry-run diffs before writes
- Deletes move to trash
- PTY disabled in daemon mode
Report for CI
Configuration quality as an exit code.
$ npx agentconfiging report --pretty
{
"findings": [...],
"stats": { ... },
"localOnly": false
}0no warning or error
1warning found
2error found
3engine failure
64usage error
FAQ
Before you run it.
Does anything leave my machine?
No. The product binds to localhost, uses a per-session token, and has no account, telemetry, or hosted component.
Do I need an account or API key?
No. Run the command in a repository. There is no signup and no product API key.
What runtimes does it detect?
Claude Code, Cursor, GitHub Copilot, Codex, Continue, Aider, Gemini CLI, and opencode. Seven additional instruction formats are available as sync targets.
Is it safe to let it write config?
Writes are path-guarded and dry-run first. You review a unified diff before committing, and deletes go to trash.
Why might search or the terminal be unavailable?
Those two features use optional native modules. The core inspector, editors, sync, catalog, and report workflow still work without them.
Can I use it in CI?
Yes. agentconfiging report emits structured JSON and returns severity-based exit codes.
What does this site collect?
Nothing. It has no analytics scripts, cookies, forms, accounts, or server-side code.