escrever-trd

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection by ingesting data from untrusted sources without explicit sanitization or boundary markers.
  • Ingestion points: The skill reads project files (Passo 1) such as source code, package.json, and configuration files. It also fetches data from external web searches and documentation MCPs (Passo 2).
  • Boundary markers: The instructions do not specify any delimiters or instructions for the agent to ignore potentially malicious embedded content in the files or search results.
  • Capability inventory: The skill has the ability to write files to the local system (docs/trd.md, docs/adrs/) and execute network-enabled tools (WebSearch, MCPs).
  • Sanitization: There is no evidence of sanitization or validation performed on the ingested content before it is processed into the final documentation templates.
  • [COMMAND_EXECUTION]: The skill invokes external tools and Model Context Protocol (MCP) servers (e.g., WebSearch, context7) to enrich documentation. This involves executing external capabilities based on parameters inferred from the local project environment.
  • [DATA_EXFILTRATION]: The skill explicitly scans files frequently associated with sensitive data, such as .env.example, .env.sample, and docker-compose.yml, searching for environment variables and API keys (e.g., STRIPE_SECRET_KEY, REDIS_URL) to identify project dependencies. While this is used for documentation purposes, it involves automated discovery of sensitive configuration patterns.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 11:13 PM
Security Audit — agent-trust-hub — escrever-trd