The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user responses from the AskQuestion tool to generate plan documents that include executable bash 'exit gates'. Malicious directives within user requirements could potentially be interpolated into these executable sections.
Ingestion points: User input from AskQuestion is used to define plan goals, subtasks, and verification logic.
Boundary markers: Uses Markdown and YAML structures, but does not explicitly instruct the agent to sanitize or ignore embedded instructions within user-provided requirements.
Capability inventory: Uses the CreatePlan tool to write .plan.md files which contain runnable bash snippets.
Sanitization: Mandates plain ASCII for YAML compatibility, which prevents some injection types, but lacks formal logic sanitization for generated scripts.
[COMMAND_EXECUTION]: The technical plan template incorporates bash code blocks for 'Exit gates' that are intended to be executed to verify plan phases. The skill also requires running git rev-parse and rg (ripgrep) to verify codebase claims during plan creation.
Evidence: references/technical-plan-template.md includes templates for runnable verification commands like npx tsc --noEmit.
Evidence: references/verification-requirements.md details how to use rg and git commands to populate verification tables.
[DYNAMIC_EXECUTION]: The skill facilitates the generation of script-like content (bash gates) based on templates and user input.
Evidence: SKILL.md and the Technical Plan Template define a process for generating file-level change instructions and associated shell-based verification logic.

tl-agent-plan-create