create-context

Pass

Audited by Gen Agent Trust Hub on Apr 2, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes local bash scripts (scripts/scan-in-folder.sh and scripts/validate-manifest.sh) via the Bash tool to automate file scanning, token estimation, and manifest validation. These scripts use standard utilities like find, stat, and wc.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from user-provided files and incorporates them into the agent's context baseline.
  • Ingestion points: Files in the .in/ directory, processed via Glob and scripts/scan-in-folder.sh.
  • Boundary markers: Missing markers to delimit or escape user-provided content in the generated CONTEXT-baseline-llm.md file.
  • Capability inventory: Access to Bash, Write, Read, and Task tools across all scripts allows for high-impact actions if the agent is influenced by malicious instructions in the source files.
  • Sanitization: The skill filters for specific sensitive file names (e.g., .env, credentials) but does not sanitize or validate the content within the ingested files.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 2, 2026, 06:18 AM
Security Audit — agent-trust-hub — create-context