The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The references/openai-agents-sdk.md file contains a code reference for managing OAuth tokens that targets the sensitive local credential path ~/.pi/agent/auth.json. It also includes a hardcoded client_id for a specific application integration.
[DATA_EXFILTRATION]: The skill provides an implementation pattern for token refreshing that reads stored authentication data from disk and transmits it to the external endpoint https://auth.openai.com/oauth/token via a network request.
[COMMAND_EXECUTION]: The skill performs shell command execution for verification purposes, specifically running npx tsc for TypeScript projects and python -c to verify library imports in Python environments. It also uses subprocess spawning to interface with the Pi CLI bridge.
[EXTERNAL_DOWNLOADS]: The skill fetches the latest project dependencies from standard registries (npm and PyPI) and retrieves real-time documentation from official Anthropic, OpenAI, and Cursor sources to ensure code generation matches current SDK versions.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it dynamically fetches and processes documentation from external websites at runtime to guide agent implementation.
Ingestion points: Documentation URLs are retrieved via WebFetch and WebSearch tools as specified in SKILL.md.
Boundary markers: No explicit boundary markers or delimiters are used in the prompts when interpolating external content.
Capability inventory: The skill has the capability to write files, install software packages, and execute shell commands.
Sanitization: There is no evidence of filtering or sanitization of the external content before it is processed for code generation.

agents-sdk-dev