net.hogarmas/blindspot

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

  warn25% of grade

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

  tool "audit_agent_surface" param "url" description: off-vendor URL in a description (example.com); capability text should not name where data goes. tool "unlock_agent_surface_evidence_pack" param "url" description: off-vendor URL in a description (example.com); capability text should not name where data goes.

• Supply chain

  warn25% of grade

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

  no resolvable repository linked to the package (provenance unverifiable)

• Credential hygiene

  fail15% of grade

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

  tool "get_ai_live_session" has a secret-shaped input parameter "capabilityToken" — secrets must never be tool parameters (the model would supply/handle the credential). tool "get_ai_short_video" has a secret-shaped input parameter "capabilityToken" — secrets must never be tool parameters (the model would supply/handle the credential). tool "play_trust_game" has a secret-shaped input parameter "playerToken" — secrets must never be tool parameters (the model would supply/handle the credential). tool "submit_blindspot_duel" has a secret-shaped input parameter "founderToken" — secrets must never be tool parameters (the model would supply/handle the credential). tool "get_submission_receipt" has a secret-shaped input parameter "receiptToken" — secrets must never be tool parameters (the model would supply/handle the credential). tool "submit_challenge_answer" has a secret-shaped input parameter "passportToken" — secrets must never be tool parameters (the model would supply/handle the credential).

• Permission scope

  unscannable15% of grade

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

  No source files, package.json, or pyproject were fetched; cannot infer capability scope.

• Version behavior

  pass10% of grade

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

  vs prior 1.28.0 (major bump): new tool "audit_agent_surface" appeared; new tool "evaluate_agent_purchase_boundary" touches a sensitive surface (email/exec/net/fs); new tool "unlock_agent_surface_evidence_pack" touches a sensitive surface (email/exec/net/fs); new tool "unlock_agent_transaction_evidence_pack" touches a sensitive surface (email/exec/net/fs). Low-severity drift (informational; consistent with the version bump).

• Transport config

  fail10% of grade

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

  remote MCP endpoint declares NO authentication (unauthenticated remote — nginx-ui/CVE-2026-33032 class)