The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exposes an attack surface for indirect prompt injection through its web research functionality.
Ingestion points: External web content fetched via WebSearch and WebFetch tools within the researcher subagent defined in SKILL.md (Step 6b).
Boundary markers: The researcher subagent is constrained by instructions to return only a compact list of verified links and titles, preventing the ingestion of full page content into the primary context.
Capability inventory: The skill has access to Bash, Write, Edit, Task, Read, Grep, and Glob tools for file manipulation and environment discovery.
Sanitization: The skill employs a 'read-only' subagent architecture for web interactions, ensuring that untrusted external data is processed in an isolated context and requires explicit user consent before activation.
[COMMAND_EXECUTION]: The skill uses the Bash tool to perform environmental checks such as determining the current git branch and remote status, as well as detecting monorepo workspace configurations (SKILL.md, modes/plan-monorepo.md). These are read-only discovery operations standard for development tools.

scope