adamsaou/overseer
Can a weaker AI catch a stronger one that's hiding something? Black-box audits for hidden objectives in language models: a CLI, a Claude Code plugin and a GitHub Action.
ARCHETYPE
Pragmatist
A balanced, no-drama setup: some rules, some tools, nothing extreme.
CONTEXT TAX · EVERY TURN
up to ~2.8k tokens
Claude Code loads MCP tools on demand, so the real cost is usually lower.
Moderate · median rig: 2.3k · breakdown
Copy this rig
# review before running: this installs third-party code
$ npx degit adamsaou/overseer/.claude ./rig-overseer # 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 adamsaou/overseer/.claude/skills/overseer-audit .claude/skills/overseer-audit $ npx degit adamsaou/overseer/plugins/overseer/skills/audit-method .claude/skills/audit-method $ npx degit adamsaou/overseer/plugins/overseer/skills/audit .claude/skills/audit $ npx degit adamsaou/overseer/plugins/overseer/skills/check .claude/skills/check $ npx degit adamsaou/overseer/plugins/overseer/skills/guard .claude/skills/guard
This rig commits no guardrails. Here is the community baseline instead — the deny/ask rules most often found across all 16.2k rigs:
{
"permissions": {
"deny": [
"Read(./.env)",
"Read(./.env.*)",
"Read(**/.env)",
"Read(~/.ssh/**)",
"Bash(git push --force:*)",
"Read(**/*.pem)",
"Read(.env)",
"Read(~/.aws/**)",
"Read(**/*.key)",
"Read(**/.env.*)",
"Bash(git push --force*)",
"Bash(rm -rf:*)",
"Bash(rm -rf *)",
"Bash(sudo *)",
"Bash(sudo:*)",
"Bash(rm -rf /)",
"Bash(git push -f:*)",
"Read(.env.*)",
"Bash(git reset --hard*)",
"Bash(rm -rf /*)"
],
"ask": [
"Bash(git push:*)",
"Bash(git push *)",
"Bash(gh pr merge *)",
"Bash(git commit:*)",
"Bash(rm *)",
"Bash(gh pr merge:*)",
"Bash(gh pr create *)",
"Bash(git reset --hard:*)",
"Bash(npm publish:*)",
"Bash(git rebase *)"
]
}
} MCP servers (1)
| server | source | est. tokens |
|---|---|---|
| overseer | local / custom | 2.5k |
Skills (5)
Slash commands (1)
/overseer
Similar rigs
juanpa1601/stepwise
Claude Code plugin for step-by-step feature delivery: spec, implement, check, audit and close, with ADRs, tests and per-step traceability on every PR.
Pragmatist 501 tok · · Q20
mcp-box/mcpscore-plugin
Audit and fix the MCP server you're building from Claude Code, Codex or Cursor. Ask your agent to audit it and get a cited score from mcpscore, with fixes applied in your own code.
Minimalist 906 tok · · Q0
Katawwr/Fusion-First
Measure and guard AI agents. OWASP-mapped attacks, a grader whose accuracy is checked every run, and a runtime guardrail that blocks unrequested tool calls. Built for open-weight models.
Pragmatist 2.9k tok · · Q0
nrzz/claude-code-starter-kits
One command sets up a lean, safe .claude/ for your project (.NET, Node, Python, Go, Flutter, Java): CLAUDE.md under 250 tokens, permissions, test skills and a cheap test-runner subagent.
Minimalist 419 tok · · Q0