pstack-harness

Warn

Audited by Gen Agent Trust Hub on Oct 23, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPERSISTENCEEXTERNAL_DOWNLOADSDYNAMIC_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The scripts/loading-path.py script executes shell commands via subprocess.run to call git show. It incorporates a user-provided --baseline argument without strict validation, which could allow for argument injection or unauthorized file access within the repository.
  • [PERSISTENCE]: The scripts/subagents.py script is designed to install custom agent definitions ('personas') into harness configuration directories such as ~/.claude/agents/ or ~/.codex/agents/. These personas contain persistent prompt instructions that modify the behavior of future agent sessions.
  • [DYNAMIC_EXECUTION]: scripts/subagents.py dynamically renders agent instructions and configuration files (Markdown and TOML) by combining templates with data from a JSON role definition file.
  • [EXTERNAL_DOWNLOADS]: The skill's documentation in references/portable-paths.md instructs the agent to clone the vendor's GitHub repository (https://github.com/mdsmithaustin/pstack.git) to initialize a source root for fresh resources. This is a vendor-owned resource but constitutes a remote download.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 23, 2026, 01:29 AM