The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's metadata includes installation instructions to add a third-party Homebrew tap (steipete/tap/summarize) and install the summarize binary. This repository is not associated with a recognized trusted organization or well-known service.
[REMOTE_CODE_EXECUTION]: By downloading and executing a binary from an external personal repository, the skill introduces a dependency on the security and integrity of that third-party source.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core functionality of ingesting and summarizing untrusted data from URLs, YouTube transcripts, and local files.
Ingestion points: External URLs, YouTube transcripts (via Apify or direct extraction), and local files.
Boundary markers: The skill includes explicit instructions for the agent to maintain security boundaries and ignore adversarial pressure, although the CLI output itself does not use specific delimiters in the provided examples.
Capability inventory: The agent can execute the summarize command, read local files, and potentially pipe output to other tools (though explicitly instructed not to pipe to shell interpreters).
Sanitization: No automated sanitization of the untrusted text is performed before it is processed by the agent.
[DATA_EXFILTRATION]: The skill manages multiple sensitive API keys (OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY, FIRECRAWL_API_KEY, APIFY_API_TOKEN) and processes local files. While the skill instructions forbid summarizing credential files (e.g., .env, .ssh/*), the fundamental process involves sending file and URL content to external LLM providers.

summarize-hardened