The Agent Skills Directory

[COMMAND_EXECUTION]: The main skill logic in SKILL.md uses node -e to execute inline JavaScript for modifying the stelow.json file and python3 -c to process JSON data from the sem tool.
[EXTERNAL_DOWNLOADS]: The skill encourages the installation of various extensions from third-party GitHub users, including PriNova/pi-agent-codebase-workflows, nicobailon/pi-intercom, and tintinweb/pi-supervisor.
[REMOTE_CODE_EXECUTION]: The codequality-review.md file provides commands to download and execute code from a GitHub repository using pi install git:github.com/cursor/plugins.
[COMMAND_EXECUTION]: The skill performs environment discovery by probing the user's home directory for hidden configuration folders such as ~/.pi/, ~/.opencode/, ~/.claude/, and ~/.codex/ to detect the running agent harness.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface (Category 8).
Ingestion points: The skill reads external technical plans (spec-tech*.md), product specifications (spec-product*.md), and live website content via the agent_browser tool.
Boundary markers: None identified in the prompt templates or file-reading logic.
Capability inventory: The skill possesses high-privilege capabilities including shell access (bash), file system writes (write), and dynamic script execution (node, python).
Sanitization: There is no evidence of sanitization or safety delimiters for data ingested from the implementation files or browser snapshots before they are processed by the LLM.

cali-product-execution-critique