Raw scan report

Anvilterm — the complete, unedited output of the deterministic mcptrustchecker engine v1.13.0, scanned . Every finding, capability tag and score component below is exactly what the engine produced — no AI, no post-processing.

← Back to the scan page
{
  "tool": {
    "name": "mcptrustchecker",
    "version": "1.13.0",
    "methodologyVersion": "mcptrustchecker-1.13"
  },
  "target": {
    "id": "anvilterm",
    "source": {
      "kind": "package",
      "origin": "anvilterm"
    },
    "server": {
      "name": "anvilterm"
    }
  },
  "grade": "C",
  "score": {
    "score": 77,
    "threatScore": 89,
    "grade": "C",
    "band": "C",
    "categorySubtotals": {
      "injection": 9.45,
      "exfiltration": 0,
      "permissions": 0,
      "supply-chain": 1.2,
      "network": 0,
      "hygiene": 0
    },
    "vector": [
      {
        "kind": "threat",
        "ruleId": "MTC-SRC-009",
        "category": "injection",
        "severity": "medium",
        "confidence": "strong",
        "rawWeight": 9,
        "confidenceMult": 0.7,
        "diminishingFactor": 1,
        "appliedPenalty": 6.3
      },
      {
        "kind": "threat",
        "ruleId": "MTC-SRC-009",
        "category": "injection",
        "severity": "medium",
        "confidence": "strong",
        "rawWeight": 9,
        "confidenceMult": 0.7,
        "diminishingFactor": 0.5,
        "appliedPenalty": 3.15
      },
      {
        "kind": "threat",
        "ruleId": "MTC-SUP-010",
        "category": "supply-chain",
        "severity": "low",
        "confidence": "heuristic",
        "rawWeight": 3,
        "confidenceMult": 0.4,
        "diminishingFactor": 1,
        "appliedPenalty": 1.2
      },
      {
        "kind": "client",
        "term": "capability-exposure",
        "level": "critical",
        "label": "capability blast radius (critical) — client exposure if the model is manipulated",
        "appliedPenalty": 10
      },
      {
        "kind": "client",
        "term": "verification-discount",
        "level": "none",
        "label": "publisher verification (unlinked) — no provenance/repo link, but the shipped source was fully read",
        "appliedPenalty": 2
      },
      {
        "kind": "client",
        "term": "coverage-honesty",
        "level": "source",
        "label": "inspection depth (source) — how much of the target the scan could see",
        "appliedPenalty": 0
      }
    ],
    "gatesFired": [],
    "methodologyVersion": "mcptrustchecker-1.13"
  },
  "capability": {
    "level": "critical",
    "reasons": [
      "ingests untrusted external content (a prompt-injection entry point)",
      "can send data / act on an external service",
      "reads sensitive or local data",
      "has a read → egress path (a data-exfiltration surface)",
      "can execute shell commands or code",
      "untrusted input can reach code execution"
    ],
    "tags": [
      "code-exec",
      "untrusted-input",
      "sensitive-source",
      "external-sink"
    ]
  },
  "coverage": {
    "level": "source",
    "inputs": {
      "toolSurface": true,
      "implementationSource": true,
      "packageMetadata": true,
      "liveTransport": false
    },
    "caveats": [
      "Tools were statically extracted from the published source (35 recovered), not enumerated from a running server. Tool-poisoning, Unicode-smuggling, capability and toxic-flow analysis ran on this inferred surface, but a mis-parsed registration could be missed or mis-attributed, so tool-derived findings are capped below “confirmed”. To grade the real runtime surface, scan the running server: --command \"npx -y <package>\"."
    ]
  },
  "findings": [
    {
      "ruleId": "MTC-SRC-006",
      "title": "Reads a sensitive credential path or dumps the environment (lib/profiles-presets.js)",
      "category": "exfiltration",
      "severity": "high",
      "confidence": "strong",
      "description": "In the server's implementation (`lib/profiles-presets.js:54`): Reading private keys / cloud credentials, or serializing the whole environment, is a sensitive-data source that becomes exfiltration when combined with any egress. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
      "remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
      "location": {
        "kind": "server",
        "name": "lib/profiles-presets.js"
      },
      "evidence": "fig', 'gcloud-home/.config/gcloud', ], }, aws: { // AWS isolation gotcha — config + credentials honour",
      "owasp": "LLM02:2025 Sensitive Information Disclosure",
      "data": {
        "rule": "MTC-SRC-006",
        "file": "lib/profiles-presets.js",
        "line": 54,
        "nonRuntime": false
      }
    },
    {
      "ruleId": "MTC-SRC-006",
      "title": "Reads a sensitive credential path or dumps the environment (server.js)",
      "category": "exfiltration",
      "severity": "high",
      "confidence": "strong",
      "description": "In the server's implementation (`server.js:608`): Reading private keys / cloud credentials, or serializing the whole environment, is a sensitive-data source that becomes exfiltration when combined with any egress. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
      "remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
      "location": {
        "kind": "server",
        "name": "server.js"
      },
      "evidence": "\" \"\\$ANVIL_AWS_FAKE_HOME/.aws/credentials\" 2>/dev/null aws() { HOME=\"\\$ANVIL_AWS_FAKE_HOME\" __anvil_aws_inner \"\\$@\"; }",
      "owasp": "LLM02:2025 Sensitive Information Disclosure",
      "data": {
        "rule": "MTC-SRC-006",
        "file": "server.js",
        "line": 608,
        "nonRuntime": false
      }
    },
    {
      "ruleId": "MTC-SRC-006",
      "title": "Reads a sensitive credential path or dumps the environment (src/client.js)",
      "category": "exfiltration",
      "severity": "high",
      "confidence": "strong",
      "description": "In the server's implementation (`src/client.js:1903`): Reading private keys / cloud credentials, or serializing the whole environment, is a sensitive-data source that becomes exfiltration when combined with any egress. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
      "remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
      "location": {
        "kind": "server",
        "name": "src/client.js"
      },
      "evidence": "ey: 'gcloud', label: '~/.config/gcloud' }, { key: 'aws', label: '~/.aws' }, { key: 'kubectl', label: '~/.kube'",
      "owasp": "LLM02:2025 Sensitive Information Disclosure",
      "data": {
        "rule": "MTC-SRC-006",
        "file": "src/client.js",
        "line": 1903,
        "nonRuntime": false
      }
    },
    {
      "ruleId": "MTC-CAP-001",
      "title": "Tool \"terminal_run\" exposes command/code execution",
      "category": "permissions",
      "severity": "high",
      "confidence": "strong",
      "description": "Tool \"terminal_run\" appears to run shell commands or evaluate code (parameter \"command\"). Arbitrary execution driven by model input is one of the most dangerous MCP capabilities; combined with any untrusted input it becomes RCE.",
      "remediation": "Sandbox execution, allowlist commands/arguments, and never pass model output to a shell unescaped.",
      "location": {
        "kind": "tool",
        "name": "terminal_run"
      },
      "owasp": "LLM05:2025 Improper Output Handling",
      "data": {
        "tags": [
          "code-exec"
        ]
      }
    },
    {
      "ruleId": "MTC-CAP-001",
      "title": "Tool \"swarm_spawn\" exposes command/code execution",
      "category": "permissions",
      "severity": "high",
      "confidence": "strong",
      "description": "Tool \"swarm_spawn\" appears to run shell commands or evaluate code (keyword \"spawn\" in tool name). Arbitrary execution driven by model input is one of the most dangerous MCP capabilities; combined with any untrusted input it becomes RCE.",
      "remediation": "Sandbox execution, allowlist commands/arguments, and never pass model output to a shell unescaped.",
      "location": {
        "kind": "tool",
        "name": "swarm_spawn"
      },
      "owasp": "LLM05:2025 Improper Output Handling",
      "data": {
        "tags": [
          "code-exec"
        ]
      }
    },
    {
      "ruleId": "MTC-SRC-002",
      "title": "Shell/command execution in server code (bin/anvil-showcase.js)",
      "category": "permissions",
      "severity": "high",
      "confidence": "strong",
      "description": "In the server's implementation (`bin/anvil-showcase.js:34151`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
      "remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
      "location": {
        "kind": "server",
        "name": "bin/anvil-showcase.js"
      },
      "evidence": "lit(\"\\n\").map((line) => exec(expandTabs(line), columns, options)).join(\"\\n\"); } // node_modules/terminal-size/index.js",
      "owasp": "LLM05:2025 Improper Output Handling",
      "data": {
        "rule": "MTC-SRC-002",
        "file": "bin/anvil-showcase.js",
        "line": 34151,
        "nonRuntime": false
      }
    },
    {
      "ruleId": "MTC-SRC-002",
      "title": "Shell/command execution in server code (bin/anvilterm-agent-loop.js)",
      "category": "permissions",
      "severity": "high",
      "confidence": "strong",
      "description": "In the server's implementation (`bin/anvilterm-agent-loop.js:24`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
      "remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
      "location": {
        "kind": "server",
        "name": "bin/anvilterm-agent-loop.js"
      },
      "evidence": "st { spawn } = require('child_process'); const { setTimeout: delay } = require('timers/promises'); // --- Arg parsing -",
      "owasp": "LLM05:2025 Improper Output Handling",
      "data": {
        "rule": "MTC-SRC-002",
        "file": "bin/anvilterm-agent-loop.js",
        "line": 24,
        "nonRuntime": false
      }
    },
    {
      "ruleId": "MTC-SRC-002",
      "title": "Shell/command execution in server code (bin/anvilterm-doctor.js)",
      "category": "permissions",
      "severity": "high",
      "confidence": "strong",
      "description": "In the server's implementation (`bin/anvilterm-doctor.js:24`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
      "remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
      "location": {
        "kind": "server",
        "name": "bin/anvilterm-doctor.js"
      },
      "evidence": "spawnSync } = require('child_process'); const ROOT = path.resolve(__dirname, '..'); const SKILL_SRC = path.join(ROOT,",
      "owasp": "LLM05:2025 Improper Output Handling",
      "data": {
        "rule": "MTC-SRC-002",
        "file": "bin/anvilterm-doctor.js",
        "line": 24,
        "nonRuntime": false
      }
    },
    {
      "ruleId": "MTC-SRC-002",
      "title": "Shell/command execution in server code (bin/anvilterm-mcp.js)",
      "category": "permissions",
      "severity": "high",
      "confidence": "strong",
      "description": "In the server's implementation (`bin/anvilterm-mcp.js:114`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
      "remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
      "location": {
        "kind": "server",
        "name": "bin/anvilterm-mcp.js"
      },
      "evidence": "st { spawn } = require('child_process'); let spawnedServerProc = null; function readSwarmMemoryForPrompt() { try {",
      "owasp": "LLM05:2025 Improper Output Handling",
      "data": {
        "rule": "MTC-SRC-002",
        "file": "bin/anvilterm-mcp.js",
        "line": 114,
        "nonRuntime": false
      }
    },
    {
      "ruleId": "MTC-SRC-002",
      "title": "Shell/command execution in server code (bin/anvilterm.js)",
      "category": "permissions",
      "severity": "high",
      "confidence": "strong",
      "description": "In the server's implementation (`bin/anvilterm.js:206`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
      "remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
      "location": {
        "kind": "server",
        "name": "bin/anvilterm.js"
      },
      "evidence": "ync, spawn } = require('child_process'); // First positional non-flag arg is the path. Empty / no arg → no-op. cons",
      "owasp": "LLM05:2025 Improper Output Handling",
      "data": {
        "rule": "MTC-SRC-002",
        "file": "bin/anvilterm.js",
        "line": 206,
        "nonRuntime": false
      }
    },
    {
      "ruleId": "MTC-SRC-002",
      "title": "Shell/command execution in server code (lib/arena-summary.js)",
      "category": "permissions",
      "severity": "high",
      "confidence": "strong",
      "description": "In the server's implementation (`lib/arena-summary.js:20`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
      "remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
      "location": {
        "kind": "server",
        "name": "lib/arena-summary.js"
      },
      "evidence": "st { spawn } = require('child_process'); const ARENA_DIR = path.join(os.homedir(), '.anvil', 'arena'); const WORKSPACES",
      "owasp": "LLM05:2025 Improper Output Handling",
      "data": {
        "rule": "MTC-SRC-002",
        "file": "lib/arena-summary.js",
        "line": 20,
        "nonRuntime": false
      }
    },
    {
      "ruleId": "MTC-SRC-002",
      "title": "Shell/command execution in server code (lib/codex-client.js)",
      "category": "permissions",
      "severity": "high",
      "confidence": "strong",
      "description": "In the server's implementation (`lib/codex-client.js:10`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
      "remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
      "location": {
        "kind": "server",
        "name": "lib/codex-client.js"
      },
      "evidence": ", execFile } = require('child_process'); const { promisify } = require('util'); const pExecFile = promisify(execFile); c",
      "owasp": "LLM05:2025 Improper Output Handling",
      "data": {
        "rule": "MTC-SRC-002",
        "file": "lib/codex-client.js",
        "line": 10,
        "nonRuntime": false
      }
    },
    {
      "ruleId": "MTC-SRC-002",
      "title": "Shell/command execution in server code (lib/forge.js)",
      "category": "permissions",
      "severity": "high",
      "confidence": "strong",
      "description": "In the server's implementation (`lib/forge.js:597`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
      "remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
      "location": {
        "kind": "server",
        "name": "lib/forge.js"
      },
      "evidence": "ecFileSync } = require('child_process'); function fetchSkillRepo(item, destDir) { if (!item.repo) throw new Error('ski",
      "owasp": "LLM05:2025 Improper Output Handling",
      "data": {
        "rule": "MTC-SRC-002",
        "file": "lib/forge.js",
        "line": 597,
        "nonRuntime": false
      }
    },
    {
      "ruleId": "MTC-SRC-002",
      "title": "Shell/command execution in server code (lib/keyring-store.js)",
      "category": "permissions",
      "severity": "high",
      "confidence": "strong",
      "description": "In the server's implementation (`lib/keyring-store.js:27`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
      "remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
      "location": {
        "kind": "server",
        "name": "lib/keyring-store.js"
      },
      "evidence": "ecFileSync } = require('child_process'); const os = require('os'); const SERVICE_PREFIX = 'anvilterm'; function fullNam",
      "owasp": "LLM05:2025 Improper Output Handling",
      "data": {
        "rule": "MTC-SRC-002",
        "file": "lib/keyring-store.js",
        "line": 27,
        "nonRuntime": false
      }
    },
    {
      "ruleId": "MTC-SRC-002",
      "title": "Shell/command execution in server code (lib/models-catalog.js)",
      "category": "permissions",
      "severity": "high",
      "confidence": "strong",
      "description": "In the server's implementation (`lib/models-catalog.js:10`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
      "remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
      "location": {
        "kind": "server",
        "name": "lib/models-catalog.js"
      },
      "evidence": "{ execFile } = require('child_process'); const { promisify } = require('util'); const pExecFile = promisify(execFile);",
      "owasp": "LLM05:2025 Improper Output Handling",
      "data": {
        "rule": "MTC-SRC-002",
        "file": "lib/models-catalog.js",
        "line": 10,
        "nonRuntime": false
      }
    },
    {
      "ruleId": "MTC-FLOW-005",
      "title": "Untrusted input can drive an external action",
      "category": "exfiltration",
      "severity": "medium",
      "confidence": "strong",
      "description": "Untrusted-input tools ([swarm_room_listen, swarm_room_thread]) co-exist with external-action tools ([terminal_run, swarm_spawn]). A prompt injection could cause unwanted external actions, though no direct sensitive-data leak path was found.",
      "remediation": "Require confirmation for state-changing/egress actions triggered after processing untrusted content.",
      "location": {
        "kind": "flow",
        "name": "swarm_room_listen → terminal_run"
      },
      "evidence": "untrusted [swarm_room_listen, swarm_room_thread] → sinks [terminal_run, swarm_spawn]",
      "owasp": "LLM06:2025 Excessive Agency",
      "references": [],
      "data": {
        "untrusted": [
          "swarm_room_listen",
          "swarm_room_thread"
        ],
        "sinks": [
          "terminal_run",
          "swarm_spawn"
        ]
      }
    },
    {
      "ruleId": "MTC-SRC-003",
      "title": "Hardcoded egress to an external endpoint (lib/models-catalog.js)",
      "category": "exfiltration",
      "severity": "medium",
      "confidence": "strong",
      "description": "In the server's implementation (`lib/models-catalog.js:75`): A hardcoded outbound call to a fixed external host inside server code is a classic exfiltration/telemetry channel — especially paired with reads of local data. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
      "remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
      "location": {
        "kind": "server",
        "name": "lib/models-catalog.js"
      },
      "evidence": "0); const r = await fetch('https://models.dev/api.json', { signal: ac.signal }); clearTimeout(t); if (!r.ok)",
      "owasp": "LLM02:2025 Sensitive Information Disclosure",
      "data": {
        "rule": "MTC-SRC-003",
        "file": "lib/models-catalog.js",
        "line": 75,
        "nonRuntime": false
      }
    },
    {
      "ruleId": "MTC-SRC-003",
      "title": "Hardcoded egress to an external endpoint (lib/usage-snapshot.js)",
      "category": "exfiltration",
      "severity": "medium",
      "confidence": "strong",
      "description": "In the server's implementation (`lib/usage-snapshot.js:58`): A hardcoded outbound call to a fixed external host inside server code is a classic exfiltration/telemetry channel — especially paired with reads of local data. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
      "remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
      "location": {
        "kind": "server",
        "name": "lib/usage-snapshot.js"
      },
      "evidence": "y { const r = await fetch('https://chatgpt.com/backend-api/wham/usage', { headers: { Authorization: `B",
      "owasp": "LLM02:2025 Sensitive Information Disclosure",
      "data": {
        "rule": "MTC-SRC-003",
        "file": "lib/usage-snapshot.js",
        "line": 58,
        "nonRuntime": false
      }
    },
    {
      "ruleId": "MTC-SRC-003",
      "title": "Hardcoded egress to an external endpoint (server.js)",
      "category": "exfiltration",
      "severity": "medium",
      "confidence": "strong",
      "description": "In the server's implementation (`server.js:4050`): A hardcoded outbound call to a fixed external host inside server code is a classic exfiltration/telemetry channel — especially paired with reads of local data. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
      "remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
      "location": {
        "kind": "server",
        "name": "server.js"
      },
      "evidence": "ll; const r = await fetch('https://api.anthropic.com/api/oauth/usage', { headers: { 'Authorization': `",
      "owasp": "LLM02:2025 Sensitive Information Disclosure",
      "data": {
        "rule": "MTC-SRC-003",
        "file": "server.js",
        "line": 4050,
        "nonRuntime": false
      }
    },
    {
      "ruleId": "MTC-SRC-009",
      "title": "Untrusted input concatenated into a command sink (bin/anvilterm-doctor.js)",
      "category": "injection",
      "severity": "medium",
      "confidence": "strong",
      "description": "In the server's implementation (`bin/anvilterm-doctor.js:55`): A shell/process command assembled from concatenated or interpolated values is command injection when any part is attacker-influenced — the OWASP canonical RCE flow. Verify what reaches the interpolated value. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
      "remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
      "location": {
        "kind": "server",
        "name": "bin/anvilterm-doctor.js"
      },
      "evidence": "n onPath(bin) { try { execSync(`command -v ${bin}`, { stdio: 'pipe' }); return true; } catch { return false; } } funct",
      "owasp": "LLM05:2025 Improper Output Handling",
      "data": {
        "rule": "MTC-SRC-009",
        "file": "bin/anvilterm-doctor.js",
        "line": 55,
        "nonRuntime": false
      }
    },
    {
      "ruleId": "MTC-SRC-009",
      "title": "Untrusted input concatenated into a command sink (bin/anvilterm.js)",
      "category": "injection",
      "severity": "medium",
      "confidence": "strong",
      "description": "In the server's implementation (`bin/anvilterm.js:323`): A shell/process command assembled from concatenated or interpolated values is command injection when any part is attacker-influenced — the OWASP canonical RCE flow. Verify what reaches the interpolated value. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
      "remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
      "location": {
        "kind": "server",
        "name": "bin/anvilterm.js"
      },
      "evidence": "'); try { execSync(`npx esbuild \"${srcPath}\" --bundle --outfile=\"${bundlePath}\" --format=iife --platform=b",
      "owasp": "LLM05:2025 Improper Output Handling",
      "data": {
        "rule": "MTC-SRC-009",
        "file": "bin/anvilterm.js",
        "line": 323,
        "nonRuntime": false
      }
    },
    {
      "ruleId": "MTC-CAP-006",
      "title": "Unconstrained command parameter \"command\" on \"terminal_run\"",
      "category": "permissions",
      "severity": "medium",
      "confidence": "heuristic",
      "description": "Tool \"terminal_run\" takes a command-shaped parameter \"command\" with no enum/pattern constraint. Free-form, model- or attacker-controlled arguments reaching a shell is the command-injection precondition.",
      "remediation": "Constrain the parameter (enum/pattern), or build the command from a fixed template with escaped args.",
      "location": {
        "kind": "tool",
        "name": "terminal_run",
        "field": "inputSchema.properties.command"
      },
      "owasp": "LLM05:2025 Improper Output Handling",
      "data": {
        "param": "command"
      }
    },
    {
      "ruleId": "MTC-SRC-005",
      "title": "Dynamic module load from a non-literal (lib/codex-client.js)",
      "category": "permissions",
      "severity": "medium",
      "confidence": "heuristic",
      "description": "In the server's implementation (`lib/codex-client.js:190`): Loading a module chosen at runtime (from a variable) can pull in and run attacker-influenced code paths. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
      "remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
      "location": {
        "kind": "server",
        "name": "lib/codex-client.js"
      },
      "evidence": "{ const mod = await import(SDK_PACKAGE); _Codex = mod.Codex; if (!_Codex) throw new Error('@openai/codex-sdk",
      "owasp": "LLM05:2025 Improper Output Handling",
      "data": {
        "rule": "MTC-SRC-005",
        "file": "lib/codex-client.js",
        "line": 190,
        "nonRuntime": false
      }
    },
    {
      "ruleId": "MTC-SRC-006",
      "title": "Reads a sensitive credential path or dumps the environment in packaging/dev tooling (public/bundle.js)",
      "category": "exfiltration",
      "severity": "low",
      "confidence": "heuristic",
      "description": "In a packaging/dev/install script (shipped, but not the server runtime) (`public/bundle.js:16979`): Reading private keys / cloud credentials, or serializing the whole environment, is a sensitive-data source that becomes exfiltration when combined with any egress. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
      "remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
      "location": {
        "kind": "server",
        "name": "public/bundle.js"
      },
      "evidence": "key: \"gcloud\", label: \"~/.config/gcloud\" }, { key: \"aws\", label: \"~/.aws\" }, { key: \"kubectl\", label: \"~/.kube\"",
      "owasp": "LLM02:2025 Sensitive Information Disclosure",
      "data": {
        "rule": "MTC-SRC-006",
        "file": "public/bundle.js",
        "line": 16979,
        "nonRuntime": true
      }
    },
    {
      "ruleId": "MTC-CAP-005",
      "title": "Mutating tool \"terminal_run\" declares no destructiveHint",
      "category": "hygiene",
      "severity": "low",
      "confidence": "heuristic",
      "description": "Tool \"terminal_run\" can mutate/egress but declares no destructiveHint. Clients that don't default to spec-safe behavior may not prompt before running it.",
      "remediation": "Declare accurate annotations, and gate destructive tools on user confirmation regardless.",
      "location": {
        "kind": "tool",
        "name": "terminal_run"
      },
      "data": {
        "tags": [
          "code-exec"
        ]
      }
    },
    {
      "ruleId": "MTC-CAP-005",
      "title": "Mutating tool \"swarm_spawn\" declares no destructiveHint",
      "category": "hygiene",
      "severity": "low",
      "confidence": "heuristic",
      "description": "Tool \"swarm_spawn\" can mutate/egress but declares no destructiveHint. Clients that don't default to spec-safe behavior may not prompt before running it.",
      "remediation": "Declare accurate annotations, and gate destructive tools on user confirmation regardless.",
      "location": {
        "kind": "tool",
        "name": "swarm_spawn"
      },
      "data": {
        "tags": [
          "code-exec"
        ]
      }
    },
    {
      "ruleId": "MTC-SRC-001",
      "title": "Dynamic code execution in packaging/dev tooling (public/lottie.min.js)",
      "category": "permissions",
      "severity": "low",
      "confidence": "heuristic",
      "description": "In a packaging/dev/install script (shipped, but not the server runtime) (`public/lottie.min.js:1`): Evaluating strings as code is the most direct RCE primitive; if any tool input reaches it, the server executes attacker-chosen code. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
      "remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
      "location": {
        "kind": "server",
        "name": "public/lottie.min.js"
      },
      "evidence": "_rt,expression_function=eval(\"[function _expression_function(){\"+val+\";scoped_bm_rt=$bm_rt}]\")[0],numKeys=property.kf?da",
      "owasp": "LLM05:2025 Improper Output Handling",
      "data": {
        "rule": "MTC-SRC-001",
        "file": "public/lottie.min.js",
        "line": 1,
        "nonRuntime": true
      }
    },
    {
      "ruleId": "MTC-SUP-010",
      "title": "Package runs install-time scripts (postinstall)",
      "category": "supply-chain",
      "severity": "low",
      "confidence": "heuristic",
      "description": "\"anvilterm\" executes postinstall script(s) at install time. An install hook runs at install time; most are routine build/setup, but review what it does before trusting it.",
      "remediation": "Review the scripts; install with --ignore-scripts where possible and vet what they do.",
      "location": {
        "kind": "package",
        "name": "anvilterm"
      },
      "evidence": "node lib/postinstall.js",
      "owasp": "LLM03:2025 Supply Chain",
      "references": [
        "https://github.com/ossf/package-analysis",
        "https://owasp.org/www-project-top-10-ci-cd-security-risks/"
      ],
      "data": {
        "scripts": [
          "postinstall"
        ]
      }
    },
    {
      "ruleId": "MTC-SUP-011",
      "title": "Package declares no source repository",
      "category": "supply-chain",
      "severity": "info",
      "confidence": "strong",
      "description": "\"anvilterm\" declares no repository URL, so its published artifact cannot be compared against reviewable source.",
      "remediation": "Prefer packages that link to public, reviewable source.",
      "location": {
        "kind": "package",
        "name": "anvilterm"
      }
    }
  ],
  "toxicFlows": [
    {
      "id": "flow-1",
      "severity": "medium",
      "confidence": "strong",
      "untrustedInput": [
        "swarm_room_listen",
        "swarm_room_thread"
      ],
      "sensitiveSource": [],
      "externalSink": [
        "terminal_run",
        "swarm_spawn"
      ],
      "selfContained": false,
      "description": "Untrusted input can drive an external action even though no sensitive source is exposed."
    }
  ],
  "capabilities": [
    {
      "tool": "terminal_create",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "terminal_write",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "terminal_run",
      "tags": [
        "code-exec"
      ],
      "reasons": {
        "code-exec": [
          "parameter \"command\""
        ]
      }
    },
    {
      "tool": "terminal_read",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "terminal_screen",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "terminal_screenshot",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "terminal_list",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "terminal_close",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "tui_type",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "tui_interrupt",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "tui_choose",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "tui_paste_ref",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "swarm_spawn",
      "tags": [
        "code-exec"
      ],
      "reasons": {
        "code-exec": [
          "keyword \"spawn\" in tool name"
        ]
      }
    },
    {
      "tool": "swarm_route",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "swarm_vendors",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "swarm_room_post",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "swarm_room_listen",
      "tags": [
        "untrusted-input"
      ],
      "reasons": {
        "untrusted-input": [
          "keyword \"fetch\""
        ]
      }
    },
    {
      "tool": "swarm_room_thread",
      "tags": [
        "untrusted-input"
      ],
      "reasons": {
        "untrusted-input": [
          "keyword \"fetch\""
        ]
      }
    },
    {
      "tool": "swarm_room_list",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "swarm_memory_read",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "swarm_memory_update",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "arena_push_artifact",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "arena_current",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "swarm_artifact_save",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "analyze_video_or_youtube",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "video_analyze",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "lookup_models",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "arena_estimate_cost",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "arena_launch",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "image_battle_launch",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "swarm_check_stuck",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "task_list",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "task_create",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "task_update",
      "tags": [],
      "reasons": {}
    },
    {
      "tool": "task_delete",
      "tags": [],
      "reasons": {}
    }
  ],
  "surfaceDigest": "f126001bce5983b0386e1a1311b44133e5cecc8aae8d374455c268f1d7021be9",
  "stats": {
    "tools": 35,
    "prompts": 0,
    "resources": 0,
    "findingsBySeverity": {
      "critical": 0,
      "high": 15,
      "medium": 8,
      "low": 5,
      "info": 1
    }
  }
}