The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external content from web pages and session transcripts, creating a surface for potential indirect prompt injection attacks.
Ingestion points: The agent is instructed to read local files within ~/Library/Caches/superpowers/browser/ (where browser session transcripts are stored) and fetch remote content from developer.apple.com and sosumi.ai using tools like WebFetch.
Boundary markers: The skill does not provide specific instructions for the agent to use delimiters or safety markers to differentiate between its instructions and the documentation it reads.
Capability inventory: The agent has access to file-reading capabilities (Read tool), network access (WebFetch), and browser controls.
Sanitization: There are no explicit instructions to sanitize or validate the content retrieved from external sources or local cache files before processing.
[COMMAND_EXECUTION]: The skill provides the agent with instructions to use local command-line tools and shell commands to manage its workflow.
Evidence: In skills/apple-docs-research.md, the agent is instructed to find session directories using ls -lt ~/Library/Caches/superpowers/browser/*/session-*. Additionally, it suggests falling back to the /opt/homebrew/bin/agent-browser CLI tool if the standard browser plugin is unavailable.
[EXTERNAL_DOWNLOADS]: The skill facilitates the retrieval of documentation from external domains to provide up-to-date technical information.
Evidence: The instructions guide the agent to fetch content from developer.apple.com (Apple's official developer site) and sosumi.ai (a markdown mirror for Apple documentation). These activities are consistent with the skill's primary purpose of documentation research.

axiom-apple-docs