The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests external data from several sources which are then used to build prompts for specialized subagents, potentially allowing malicious instructions in a task file to influence the agent's actions.
Ingestion points: Processes user-provided arguments, local task files (TASK_N.md), and project metadata (CLAUDE.md, package.json). (File: SKILL.md)
Boundary markers: No explicit delimiters or 'ignore instructions' markers are defined for the content passed into the subagent dispatch prompts.
Capability inventory: Spawns subagents via a Task tool with full repository access, executes shell commands for testing/builds, and utilizes network-enabled MCPs like firecrawl. (File: SKILL.md, references/agent-dispatch.md)
Sanitization: The skill lacks sanitization or validation of the task descriptions before they are interpolated into the execution workflow.
[COMMAND_EXECUTION]: The skill identifies and executes shell commands for quality gates and baseline verification.
Evidence: Automatically discovers and runs project-specific commands like tsc, lint, test, and build from package.json. It also uses curl to establishment baseline API response times and status codes. (File: SKILL.md, references/verification-protocol.md)
[DYNAMIC_EXECUTION]: The skill is designed to orchestrate the creation, modification, and execution of application code at runtime.
Evidence: Uses the Task tool to delegate implementation of components, hooks, routes, and services to specialized agents based on technical requirements discovered during the analysis phase. (File: SKILL.md, references/agent-dispatch.md)
[EXTERNAL_DOWNLOADS]: The skill utilizes external tools to fetch documentation and verify web pages.
Evidence: Instructs the use of the firecrawl MCP to fetch external documentation or references from the web and context7 to query library documentation for specific versions. (File: SKILL.md)

execute-task