The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to detect missing dependencies and output an installation command for the user: npx skills add FuDesign2008/open-skills -g --skill '*' --yes. This references a repository managed by the skill's author.
[COMMAND_EXECUTION]: The skill allows the use of shell commands (Bash) for read-only verification purposes, such as running the application under debug or executing reproduction steps, with a requirement for user confirmation for execution.
[INDIRECT_PROMPT_INJECTION]: The skill functions as an analysis engine for untrusted external data including source code files, logs, and user-provided descriptions. It mitigates injection risks by requiring a 'red-capable loop' (agent-verified failing state) and excluding user-pasted evidence as a primary source for root-cause hypotheses.
Ingestion points: Codebase files via Read/Grep/SemanticSearch, user-pasted logs, and web search results in SKILL.md sections 1 and 2.
Boundary markers: The skill uses a 'red-capable loop gate' to ensure the agent independently verifies symptoms rather than obeying embedded instructions in data.
Capability inventory: File system read/write (limited to instrumentation), Bash shell (read-only), and WebSearch.
Sanitization: Relies on logical verification gates rather than string sanitization.

analysis-core