The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes several Bash and Python scripts for core operations, including init-project-data.sh, maintain-project-state.sh, and doctor_engine.py. These scripts manage the project's internal data structure and lifecycle state.
[EXTERNAL_DOWNLOADS]: The run-upgrade.sh script performs a git pull or git fetch from its origin repository to update the skill's source code. Additionally, documentation in taste-review/content.md references fetching style guides from Google's official GitHub repository via curl.
[PROMPT_INJECTION]: The skill implements highly autonomous pipelines. For instance, pipeline.run-when-committing.md explicitly instructs the agent to extract and write insights and perform git commits without requesting user confirmation or showing drafts for approval.
[COMMAND_EXECUTION]: The install-hooks.sh script modifies the agent's global configuration file (~/.claude/settings.json) to register hooks that execute the skill's scripts during session starts and tool usage. This is a form of persistence used to provide continuous context injection.
[PROMPT_INJECTION]: An attack surface for indirect prompt injection exists because the skill automatically injects content from project-level markdown files (in the .pensieve/ directory) into the agent's prompt during 'Explore' and 'Plan' tasks. Malicious instructions placed in these files by an external actor could influence the agent's behavior when the skill is active.

pensieve