me.openbrowser/openbrowser-ai

What we checked

Six static checks, weighted by risk. Every result reflects only what could be observed in the published package and repository — never intent.

• Injection surface

  unscannable25% of grade

  Tool descriptions/manifest scanned for instruction-injection patterns (imperatives at the model, hidden text, 'ignore previous', data-exfil URLs).

  No tool descriptions, server instructions, prompts, or resources were fetched; nothing to scan for injection.

• Supply chain

  fail25% of grade

  Package provenance: namespace verification, repo linkage, maintainer count, account age, postinstall scripts, typosquat distance.

  Python build script executes at build time: network call in build script

• Credential hygiene

  warn15% of grade

  How the server takes secrets (env vs plaintext config vs hardcoded); secrets appearing in tool schemas.

  frontend/src/components/browser/BrowserViewer.tsx: assigns a literal to password (${vn…d} (len 19)) — non-provider secret literal; prefer environment intake. src/openbrowser/config.py: assigns a literal to api_key (your…re (len 24)) — non-provider secret literal; prefer environment intake. Reads secrets from the environment (e.g. .opencode/plugins/openbrowser.js) — the recommended intake shape.

• Permission scope

  fail15% of grade

  Declared tools vs. breadth (filesystem, network, exec); flags shell-exec and unbounded filesystem access.

  HIGH: downloads and pipes straight to a shell (infra/eval/terraform/user_data.sh: pipes a download into a shell). has both shell/exec and destructive filesystem writes (benchmarks/e2e_llm_benchmark.py: subprocess.* call; benchmarks/e2e_llm_benchmark.py: open(…, 'w'/'a'/'x') write-mode file access). discloses shell/exec capability: 13 files (benchmarks/e2e_llm_benchmark.py, examples/apps/ad-use/ad_generator.py, infra/eval/pipelines/eval_benchmark.py, +10 more) discloses dynamic code execution: 15 files (infra/training/finetuning/eval_sft.py, infra/training/finetuning/grpo_trainer.py, infra/training/finetuning/online_grpo_trainer.py, +12 more) discloses broad/destructive filesystem access: 26 files (benchmarks/e2e_llm_benchmark.py, benchmarks/e2e_llm_stats.py, examples/apps/ad-use/ad_generator.py, +23 more) Capability DISCLOSURE, not a verdict: static analysis sees the primitive is present and reachable, not whether its use is attacker-controlled.

• Version behavior

  unscannable10% of grade

  Diff of tool definitions between versions; new permissions or changed descriptions in a patch release (the postmark-mcp class).

  no prior version to diff (first sight of me.openbrowser/[email protected]); version-drift / rug-pull cannot be evaluated. Cold-start risk is covered by point-in-time checks (injection_surface, supply_chain, credential_hygiene), not here.

• Transport config

  fail10% of grade

  Remote servers: TLS and auth mode (none/token/OAuth). Local servers: whether the manifest indicates it phones home.

  source performs runtime self-update / download-and-execute (bypasses version_behavior); sends telemetry to third-party SDK (posthog) — disclosure, verify opt-in; manifest/source references telemetry behavior — disclosure