AGENT SKILL

trust-boundary-analysis

Trust boundary mapping and startup sequence audit for developer tools, CLI apps, and plugin systems. Load when the target is a developer tool, CLI, IDE extension, or any application that loads config from the current directory.

License 许可
Apache-2.0
所属仓库
Akapi895/raptix
最近更新
2026-11-06
Stars
★ 0
Forks
0

01 前言 Front Matter

name
trust-boundary-analysis
metadata
{ "subdomain": "analyst", "when_to_use": "trust boundary startup sequence developer tool cli ide extension plugin config load current directory analysis", "mitre_attack": "T1574" }
description
Trust boundary mapping and startup sequence audit for developer tools, CLI apps, and plugin systems. Load when the target is a developer tool, CLI, IDE extension, or any application that loads config from the current directory.

03 SKILL.md 正文

Trust Boundary Analysis

This skill targets the vulnerability class that produced 5 RCE vectors in Google Gemini CLI from a single architectural flaw: missing workspace trust. Developer tools that auto-load configuration from untrusted directories are a rich attack surface.

When to Use

Apply this skill when the target:

  • Loads .env, settings.json, or config files from the current working directory
  • Has a plugin/extension system that auto-discovers and loads code
  • Spawns child processes with configuration-controlled commands
  • Is a CLI tool, IDE extension, language server, or build tool
  • Uses MCP (Model Context Protocol) servers configured via project files

Startup Sequence Audit

For any CLI tool or developer application, trace the full initialization:

1. Config file discovery

grep -rn 'readFile\|readFileSync\|fs.read\|open(' --include='*.ts' --include='*.js' \
  /workspace/target/src/ | grep -i 'config\|settings\|env\|rc\|\.json'

Map the search order. Common dangerous patterns:

  • cwd/.tool/config.json → cwd/.env → ~/.tool/config → /etc/tool/config
  • If the local (cwd) config is loaded BEFORE the user's global config, the attacker's repo-level config wins.

Record each config loading point:

kg_record('[{"kind": "CodeLocation", "key": "code-location::loadConfig-cwd-env",
  "label": "loadConfig() reads .env from cwd",
  "props": {"file": "src/config/settings.ts", "line": 42,
            "trust_level": "untrusted"}}]')

2. Environment variable injection

grep -rn 'process\.env\|os\.environ\|env::var\|getenv' --include='*.ts' \
  --include='*.py' --include='*.rs' /workspace/target/src/

Check: Are env vars from .env files injected into process.env? Which vars control dangerous behavior? Look for:

  • *_COMMAND, *_CMD, *_EXEC → shell execution
  • *_PROXY → SSRF / network interception
  • *_PATH, *_DIR → path traversal
  • *_URL → open redirect / SSRF
  • DEBUG, NODE_ENV → bypass security controls

3. Workspace trust check

grep -rn 'trust\|isTrusted\|workspace.*safe\|folder.*trust' --include='*.ts' \
  --include='*.js' /workspace/target/src/

Look for the pattern:

// DANGEROUS: trust disabled by default
if (!settings.folderTrustEnabled) {
  return { isTrusted: true };  // ← All workspaces trusted
}

If no trust check exists, every config is loaded blindly. This is the root cause.

4. Command execution from config

grep -rn 'spawn\|exec\|execSync\|child_process\|subprocess\|os\.system\|Popen' \
  --include='*.ts' --include='*.js' --include='*.py' /workspace/target/src/

For each spawn/exec call, check:

  • Is shell: true set? → shell injection via crafted values
  • Does the command come from config? → attacker-controlled execution
  • Is there any validation/allowlist on the command?

5. Plugin/tool auto-discovery

grep -rn 'discoverTools\|loadPlugins\|autoDiscover\|mcpServers\|toolDiscovery' \
  --include='*.ts' --include='*.js' /workspace/target/src/

Check: Are plugins/tools loaded from project-level config BEFORE any user confirmation? MCP servers, language server plugins, and build tool extensions are common vectors.

Trust Boundary Graph Modeling

Map the attack flow as a chain in the knowledge graph:

ENTRYPOINT (malicious .env file in cloned repo)
  → enables → ENV_INJECT (process.env poisoned)
    → enables → SHELL_EXEC (spawn with shell:true)
      → reaches → CROWN_JEWEL (arbitrary code execution)

Use low edge weights (0.2–0.4) for automatic/silent steps. Use higher weights (1.0+) for steps requiring user interaction.

Then record the chain as an AttackPath observation via kg_record; chain planning is not on your tool surface.

Shell Quoting Vulnerabilities

When config values flow into shell commands, check for:

  1. Direct interpolation: `command ${userValue}` → injection via ; id
  2. Single-quote escape: osascript -e '${path}' → escape via ' in path
  3. Env var in shell context: PROXY_CMD="curl evil.com" → executed as-is
  4. Argument injection: spawn('git', ['clone', userUrl]) → --upload-pack= prefix

For each, the PoC pattern is:

  • Craft input that breaks out of the quoting context
  • Inject a command that proves execution (id, env, write a sentinel file)
  • Use validate_finding with the sentinel as the success pattern

Negative Control

Always test: does the same code path execute WITHOUT the malicious config? The negative control is the same operation with a clean/empty config directory. If the behavior occurs regardless, it's not a vulnerability — it's expected.

04 同仓库更多 Skills