The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a multi-step research and planning workflow that ingests external data. 1. Ingestion points: The workflows/research-phase.md workflow uses web search tools to fetch data which is then stored in FINDINGS.md. 2. Boundary markers: The skill uses structured XML tags like <task>, <action>, and <objective> in its templates to delimit instructions. 3. Capability inventory: The workflows/execute-phase.md workflow has the capability to execute shell commands and tool calls defined in PLAN.md files. 4. Sanitization: The workflow does not specify explicit programmatic sanitization or filtering of the external content retrieved during the research phase.
[DYNAMIC_EXECUTION]: The skill's primary function involves generating PLAN.md files based on project briefs and roadmaps, which are subsequently parsed and executed as instruction sets. The workflows/execute-phase.md file describes a process for routing these instruction segments to subagents or the main agent context for execution, representing dynamic execution of generated content.
[DYNAMIC_CONTEXT_INJECTION]: The skill utilizes the dynamic context injection syntax !date +%Y-%m-%d within the templates/research-prompt.md file to fetch the current system date. This is a benign use of the platform's features for environment-aware templating.
[COMMAND_EXECUTION]: The references/cli-automation.md reference file explicitly guides the agent to automate a variety of third-party command-line interfaces for services like Vercel, Stripe, and GitHub, facilitating extensive interaction with the local shell environment as part of the implementation workflow.

create-plans