everything-claude-code-harness

Warn

Audited by Gen Agent Trust Hub on Sep 12, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The installation instructions direct the user to clone a repository and execute a shell script (./install.sh) with arguments. This pattern facilitates the execution of arbitrary remote code on the host system.
  • [EXTERNAL_DOWNLOADS]: The skill fetches resources from external sources, including a GitHub repository (github.com/affaan-m/everything-claude-code) and npm packages (ecc-universal, ecc-agentshield).
  • [COMMAND_EXECUTION]: The skill utilizes several slash commands and Node.js utility scripts (e.g., setup-package-manager.js) to modify environment variables, project configurations, and global settings.
  • [PERSISTENCE]: The skill implements lifecycle hooks (hooks/session-start.js, hooks/stop.js) that read and write session data to ~/.claude/session-memory.json. This maintains agent state and context across different sessions.
  • [INDIRECT_PROMPT_INJECTION]:
  • Ingestion points: The skill processes user prompts, session summaries (ECC_SESSION_SUMMARY), and instinct files to guide agent behavior.
  • Boundary markers: None identified in the provided implementation snippets.
  • Capability inventory: Includes execution of shell scripts, management of package managers, and file system writes to memory stores.
  • Sanitization: There is no evidence of sanitization for session summaries or instincts before they are persisted or re-injected into the context.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 12, 2026, 08:17 PM
Security Audit — agent-trust-hub — everything-claude-code-harness