FlorianBruniaux/ctxharness
Detect drift in AI instruction files, audit Claude Code and Codex configuration, and export portable reports.
ARCHETYPE
Fort Knox
Deny lists, pre-tool hooks, sandboxing. Nothing touches prod without a signature.
CONTEXT TAX · EVERY TURN
up to ~7.2k tokens
Claude Code loads MCP tools on demand, so the real cost is usually lower.
Heavy · median rig: 2.3k · breakdown
GUARDRAILS
4/5
Blocks destructive commands · Protects secrets · Pre-tool screening hook · No YOLO mode · details
Quality 55/100 Guardrails v2
Copy this rig
# review before running: this installs third-party code
$ npx degit FlorianBruniaux/ctxharness/.claude ./rig-ctxharness # inspect, then merge into .claude/ MCP servers are added to Claude Code at local scope; env vars are shown as YOUR_… placeholders — we never store values. Files are fetched with degit into a separate folder so you can review before merging.
$ npx degit FlorianBruniaux/ctxharness/.claude/skills/eval-rules .claude/skills/eval-rules $ npx degit FlorianBruniaux/ctxharness/.claude/skills/eval-skills .claude/skills/eval-skills $ npx degit FlorianBruniaux/ctxharness/.claude/skills/token-audit .claude/skills/token-audit $ npx degit <redacted>-config/__fixtures__/claude-project/.claude/skills/duplicate-review .claude/skills/duplicate-review $ npx degit <redacted>-config/__fixtures__/claude-project/.claude/skills/portable-demo .claude/skills/portable-demo $ npx degit <redacted>-config/__fixtures__/profile-expectations/sources/A/.claude/skills/demo .claude/skills/demo
$ curl -fsSL --create-dirs -o .claude/agents/reviewer.md https://raw.githubusercontent.<redacted>-config/__fixtures__/claude-project/.claude/agents/reviewer.md Merge into .claude/settings.json (project) or ~/.claude/settings.json (user). Hook commands reference scripts in the source repo — copy those too.
{
"permissions": {
"deny": [
"Bash(cat .env*)",
"Read(**/.env*)",
"Read(**/*.pem)",
"Read(**/*.key)",
"Read(**/*.cert)",
"Read(**/node_modules/**)"
]
},
"hooks": {
"PreToolUse": [
{
"matcher": "Bash",
"hooks": [
{
"type": "command",
"command": "bash .claude/hooks/dangerous-actions-blocker.sh"
},
{
"type": "command",
"command": "bash .claude/hooks/rtk-suggest.sh"
}
]
},
{
"matcher": "Write|Edit",
"hooks": [
{
"type": "command",
"command": "bash .claude/hooks/dangerous-actions-blocker.sh"
},
{
"type": "command",
"command": "bash .claude/hooks/security-gate.sh"
}
]
}
]
}
} MCP servers (2)
| server | source | est. tokens |
|---|---|---|
| placeholder env: MCP_TOKEN | local / custom | 2.5k |
| literal env: API_KEY | local / custom | 2.5k |
Skills (6)
Subagents (1)
| reviewer | Review fixture changes without writing files. |
Hooks (11)
| event | matcher | runs |
|---|---|---|
| PreToolUse | Bash | bash .claude/hooks/dangerous-actions-blocker.sh |
| PreToolUse | Bash | bash .claude/hooks/rtk-suggest.sh |
| PreToolUse | Write|Edit | bash .claude/hooks/dangerous-actions-blocker.sh |
| PreToolUse | Write|Edit | bash .claude/hooks/security-gate.sh |
| PostToolUse | Write|Edit | bash .claude/hooks/auto-format.sh |
| PostToolUse | Write|Edit | bash .claude/hooks/typecheck-feedback.sh |
| UserPromptSubmit | * | bash .claude/hooks/git-context.sh |
| UserPromptSubmit | * | bash .claude/hooks/smart-suggest.sh |
| UserPromptSubmit | * | ./.claude/hooks/check.sh |
| UserPromptSubmit | * | ./.claude/hooks/missing.sh |
| UserPromptSubmit | * | fixture-hook-from-path |
Slash commands (3)
/audit-docs/commit/release
Permissions
deny (6)
Bash(cat .env*)
Read(**/.env*)
Read(**/*.pem)
Read(**/*.key)
Read(**/*.cert)
Read(**/node_modules/**)
ask (0)
—
allow (10)
Bash(pnpm *)
Bash(node *)
Bash(git *)
Bash(npm *)
Bash(gh *)
Bash(find *)
Bash(ls *)
Bash(cat *)
Bash(chmod *)
Bash(rtk *)
Similar rigs
LISAooo1234/Bunny-Byte
—
Minimalist 2.0k tok · · Q0
OMOCHInoHOSHI/dotfiles-claude
.claude
Minimalist 177 tok · · Q19
Clarittyai/agentic-app-seed
Self-hostable template for building AI workflow automations and agentic apps: FastAPI, React and Postgres, wired for Claude Code. Docker-ready.
Pragmatist 11.3k tok · · Q25
MG-Cafe/claude-code-best-engineers-setup
Nine practices from the team that builds Claude Code: a fresh reviewer agent, ask before any push, hooks that fail closed, reviews that finish, effort per subagent and more. The setup from my YouTube video.
Fort Knox 146 tok · · Q35