{
  "markdown": "# arrakis-careers\n\nTwo commands. See what your agents expose, then apply to Arrakis Security from\nyour terminal.\n\nWe do not run a normal recruiting funnel. No ATS, no résumé upload, no\nseven-round loop. Install this, run two commands, done in under two minutes.\n\n## Install\n\n**Claude Code**\n\n```\n/plugin marketplace add Arrakis-Security/careers\n/plugin install arrakis-careers@arrakis\n```\n\n**Codex**\n\n```bash\ncodex plugin marketplace add Arrakis-Security/careers\ncodex plugin add arrakis-careers@arrakis\n```\n\n**Antigravity**\n\n```bash\nagy plugin install https://github.com/Arrakis-Security/careers\n```\n\nThe Claude Code command is verified against this repository — not a local\ncheckout — on 2026-08-06, Claude Code 2.1.143. The Codex and Antigravity\ncommands come from testing against a local checkout and have not been re-run\nagainst the public repository. We would rather name the gap than let a command\nthat returns \"plugin not found\" be your first impression. If one fails for you,\nmail build@arrakis.security and we will fix it fast.\n\nNo dependencies. Nothing to build. Nothing runs at install time.\n\n## Use\n\nIn Claude Code, the commands are namespaced:\n\n```\n/arrakis-careers:scan     # what your agents expose. read-only, sends nothing.\n/arrakis-careers:apply    # three questions, no résumé.\n```\n\nIn Codex and Antigravity, just ask — \"scan my agent surface\" or \"apply to\nArrakis\" — and the skill fires.\n\n`scan` inspects your own machine: which MCP servers your agents are configured\nwith, what those servers can reach, how broad the credential and filesystem\nscopes are, and which tools ingest untrusted content and could carry a prompt\ninjection. Ranked, redacted, and printed — nowhere else.\n\nThen it makes the point we are hiring for: that is one laptop. Our customers run\nthousands of agents against production systems.\n\n`apply` asks three questions. What you broke or built recently, a link, and how\nto reach you. Optionally which of our four domains you are aiming at.\n\nSubmission is not switched on yet, and we would rather say so than let you think\nyou had applied. `apply` writes your answers to a file in the current directory\nand hands you a pre-filled mail to build@arrakis.security. Send it and someone\nreads it.\n\n## Security\n\n`scan` sends nothing to Arrakis and never prints a secret value. The only thing\never sent anywhere is your application, after you approve the exact payload on\nscreen. Scan findings are never attached to it.\n\nOne thing worth knowing before you run it: this plugin is instructions, not code,\nso the config files `scan` reads pass through your own model provider the same\nway any file you hand your agent does. [SECURITY.md](SECURITY.md) explains\nexactly what that means, what we do and do not see, and gives you a two-minute\nway to check the whole thing yourself.\n\n## Why you can trust what you install\n\nYou are being asked to run instructions against your own credential stores by a\ncompany you may not know. That deserves infrastructure, not assurances.\n\n- **Nobody outside Arrakis Security can approve a change here.** Every path is\n  owned in [.github/CODEOWNERS](.github/CODEOWNERS), code-owner review is\n  required on the default branch, and the ruleset has no bypass list. Outside\n  contributions are welcome and arrive as pull requests from forks.\n- **Every pull request is machine-checked before a human sees it.**\n  [scripts/check_content_safety.py](scripts/check_content_safety.py) fails any new\n  line in the prompt surface that touches a network verb, a secret path, a shell,\n  or an instruction override until a maintainer records that exact line. It also\n  rejects invisible characters, lookalike letters, and base64 blobs — the shapes a\n  malicious contribution takes in a repository made of prose.\n- **`SUBMIT_ENABLED = false` is enforced by CI**, not by convention.\n- **Signed commits, linear history, signed release tags**, and installs you can\n  pin to a tag: [docs/RELEASING.md](docs/RELEASING.md).\n- **The threat model is written down**, including the attacks we think are most\n  likely against this repository specifically:\n  [docs/THREAT-MODEL.md](docs/THREAT-MODEL.md).\n\nRun our own guardrail on your clone before you install, if you like:\n\n```bash\ngit clone --depth 1 https://github.com/Arrakis-Security/careers && cd careers\npython3 scripts/check_content_safety.py\n```\n\n## Contributing\n\nPull requests welcome, including from outside. Read\n[CONTRIBUTING.md](CONTRIBUTING.md) first — the constraint is that markdown here is\nexecutable, so review is line by line and only staff can merge.\n\nSecurity defects go to\n[private reporting](https://github.com/Arrakis-Security/careers/security/advisories/new)\nor build@arrakis.security, never a public issue.\n\n## Who we are\n\nArrakis Security secures the autonomous workforce. Agent discovery,\npre-execution policy enforcement, an MCP gateway, behavioural anomaly detection,\nand kill-switches — for enterprises whose AI agents are already inside core\nbusiness systems.\n\nWe are hiring across four domains, listed in\n[skills/shared/arrakis.md](skills/shared/arrakis.md). Software engineers and\ndata scientists first.\n\nbuild@arrakis.security\n",
  "bytes": 5148,
  "sha": "5d667dfcc15a5cb2de57fd1c15460b49b8e18727cbfd53e6e56d47c2b6cde144",
  "repo_slug": "arrakis-security/careers",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/plg_arrakis_security_careers_4b2f80e4/readme"
}