Back to skill

Security audit

Google Health

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious because the scanners report a risk of authenticated Google requests being redirected, configurable OAuth scopes that could exceed the stated read-only purpose, and exercise data omissions that conflict with its description.

Consider the suspicious verdict and the reported possibility of redirected authenticated requests and broader OAuth scopes before installing. The scan is partial: although all listed components were inspected, some referenced paths were missing, so those references could not be resolved.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T09 · Insecure Skill Coding Practices

Error
Location
internal/config/config.go
Finding

Automatically discovered working-directory configuration can redirect authenticated requests

Content
View full analysis

Impact

The attacker-controlled endpoint receives the Google access token in the authenticated request and can replay it against Google for the token's granted scopes, exposing the user's health and profile data until the token expires or is revoked. This path does not demonstrate disclosure of the refresh token or OAuth client secret.

Vulnerability Patterns
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (8)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The description explicitly states that the client performs no filtering. However, buildSessionRows uses if s.Start == nil { continue }, excluding those records from both parsed JSON and human-readable output. Raw output preserves the retrieved points. The remaining visible behavior is consistent with a read-only exercise-session client; no unrelated resource access or writes are shown.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
92% confidence
Finding

resolveWindow returns filtered=true unless --all is supplied, and runDataList passes that flag and the time bounds to ListDataPoints. The error handling also explicitly identifies this as server-side time filtering. Thus, the no-filtering claim is inaccurate as written; it could instead state that no client-side filtering or merging occurs. The shown code otherwise matches the read-only listing purpose and contains no visible API writes or unrelated resource access.

Content

No source excerpt is available for this finding.

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · internal/auth/tokencache.go (reported line 118)May include surrounding context.

go
// persistingSource wraps a refreshing TokenSource and re-saves the token to disk
// whenever .Token() yields a newer one (an access-token rotation after a
// refresh). x/oauth2 handles refresh-token rotation internally; we only need to
// persist the result so the on-disk access token stays fresh between runs
// (GOAL.md §8 — ghealth doesn't do this; we should).
type persistingSource struct {
	base   oauth2.TokenSource

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · internal/auth/tokencache.go (reported line 142)May include surrounding context.

go
}

// changed reports whether tok differs from the last persisted token in a way
// worth writing (a new access token or a new expiry).
func (p *persistingSource) changed(tok *oauth2.Token) bool {
	if p.last == nil {
		return true

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · internal/cli/auth.go (reported line 119)May include surrounding context.

go
report := authStatusReport{TokenPath: cfg.TokenCache}
			if tok != nil {
				report.Present = true
				report.Valid = tok.Valid() // non-empty access token AND not expired.
				if !tok.Expiry.IsZero() {
					report.Expiry = tok.Expiry.UTC().Format(time.RFC3339)
				}

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: file_write, network, shell.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

Parsed session output silently filters out exercise points without a parseable start time.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The manifest explicitly promises no filtering and leaves data-selection decisions to the caller. In internal/cli/sessions.go, L129-L130 discard any parsed session whose Start is nil, so normal and --json output can omit retrieved exercise points; this also contradicts the no-filtering statement at L021-L024. The --raw branch preserves points, but does not make parsed output unfiltered.

Content

No source excerpt is available for this finding.

OAuth scope configuration accepts arbitrary scopes despite the read-only contract.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

applyFile replaces the read-only defaults with arbitrary configured OAuth scopes, so the claimed read-only authorization boundary is not enforced. An altered configuration could make a subsequent login request broader privileges; obtaining them still requires provider acceptance and user consent, but any resulting cached credential could exceed the tool's promised access.

Content

No source excerpt is available for this finding.