Back to skill

Security audit

thumbtack-mcp

Security checks for vulnerabilities and agentic risk

Overview

This skill reads public Thumbtack listings as advertised, with a minor risk of overwriting local intermediate files.

Run recipes in a fresh temporary directory to avoid overwriting existing JSON files. Searches send the chosen trade and ZIP code to Thumbtack; no login credentials are needed. All bundled files were reviewed to investigate scanner gaps, but live endpoint behavior and host permissions were not tested.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (11)

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · SKILL.md (reported line 3)May include surrounding context.

md
---
name: thumbtack
description: "Query thumbtack.com from a shell — search local service pros by trade and zip, and read pro profiles with ratings, reviews, credentials, business hours and pricing. Uses plain curl against Thumbtack's SSR pages and its anonymous GraphQL endpoint; no API key, no login, no browser extension. Use when you want Thumbtack data without an MCP server, in a script, or one-shot."
---

# Thumbtack access (curl)

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · SKILL.md (reported line 44)May include surrounding context.

Read a pro profile — the ld+json path is simplest:

sh
curl -sSL -A "$UA" "<profile-url>" | node "$TT" ldjson | jq '.["@graph"][0]'

For credentials, business hours, payment methods and specialties use

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · references/discovery.md (reported line 16)May include surrounding context.

Thumbtack canonicalises loose slugs by redirect. Ask it rather than guessing:

sh
curl -sSL -o /dev/null -A "$UA" -w '%{url_effective}\n' \
  "https://www.thumbtack.com/k/plumbing/near-me?zip_code=28203"
# https://www.thumbtack.com/k/plumbers/near-me?zip_code=28203   <- canonical slug is "plumbers"

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · references/discovery.md (reported line 63)May include surrounding context.

Pro profile — summary (ld+json, simplest)

sh
curl -sSL -A "$UA" "$PROFILE_URL" | node "$TT" ldjson | jq -c '.["@graph"][0] | {
   name, description,
   rating: (.aggregateRating.ratingValue*100|round/100),
   reviews: .aggregateRating.reviewCount,

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · references/discovery.md (reported line 86)May include surrounding context.

review is an array of arrays, so flatten first:

sh
curl -sSL -A "$UA" "$PROFILE_URL" | node "$TT" ldjson \
 | jq -c '[.["@graph"][0].review | flatten[] | {
     stars: .reviewRating.ratingValue, by: .author.name,
     date:  .datePublished, text: .description }] | .[0:10][]'

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · references/discovery.md (reported line 97)May include surrounding context.

These are absent from ld+json. Pull the section list once:

sh
curl -sSL -A "$UA" "$PROFILE_URL" | node "$TT" apollo > apollo.json

SP='.ROOT_QUERY | to_entries | map(select(.key|startswith("servicePage")))[0].value'

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · tt-extract.mjs (reported line 3)May include surrounding context.

js
#!/usr/bin/env node
// Extract Thumbtack SSR data stores from a page on stdin.
// Usage: curl -sSL <url> | node tt-extract.mjs next|apollo|ldjson
//
// Dependency-free on purpose: this runs from a skills directory with no
// node_modules. Its semantics mirror the server's `@chrischall/mcp-utils`

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

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/discovery.md (reported line 129)May include surrounding context.

Useful for paging past the SSR page's 10 results / fixed review window.

sh
curl -sS -X POST https://app.thumbtack.com/graphql -A "$UA" \
  -H 'content-type: application/json' -H 'origin: https://www.thumbtack.com' \
  -d '{"query":"query{__typename}"}'
# {"data":{"__typename":"Query"}}

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/discovery.md (reported line 145)May include surrounding context.

jq -c '{query:"query SP($input: ServicePageInput!){ servicePage(input:$input){ __typename shareableURL hasVerifiedPhoneNumber sections{ __typename } } }", variables: .}' vars.json > q.json

curl -sS -X POST https://app.thumbtack.com/graphql -A "$UA"
-H 'content-type: application/json' -H 'origin: https://www.thumbtack.com' -d @q.json

text

Recipes overwrite fixed filenames without warning about existing-file loss.

Low
Category
Output Handling
Confidence
97% confidence
Finding

Shell redirection to fixed names such as apollo.json, vars.json, and q.json truncates existing files in the working directory without confirmation. Running these recipes in a project directory can destroy unrelated data, and preexisting symlinks can redirect writes to other files writable by the process.

Content

No source excerpt is available for this finding.