harness-persist

Pass

Audited by Gen Agent Trust Hub on Jul 15, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute a local Python script (scripts/resolve_harness.py) to determine the file system path for the Project Harness. This script is part of the accompanying harness-init toolset and is used to locate the shared directory across Git worktrees.
  • [PROMPT_INJECTION]: The skill facilitates the persistence of user-influenced or project-derived data into files that are intended to be read by future agent sessions. This creates a surface for indirect prompt injection, where malicious instructions could be stored in the harness and later executed by an agent processing those files.
  • Ingestion points: Future agent sessions are expected to read files from the <harness>/ directory (such as project.md, index.md, and session notes) to establish context.
  • Boundary markers: There are no explicit instructions for the agent to use delimiters or ignore embedded instructions when reading from or writing to the harness files.
  • Capability inventory: The skill allows for command execution (via the resolve script) and filesystem write operations; the wider agent environment likely possesses further capabilities.
  • Sanitization: The skill does not specify any validation, filtering, or sanitization protocols for the data being persisted.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 15, 2026, 02:39 AM
Security Audit — agent-trust-hub — harness-persist