The Agent Skills Directory

[PROMPT_INJECTION]: The skill's core workflow involves analyzing codebase content to generate implementation plans, which introduces a surface for indirect prompt injection (Category 8). Maliciously crafted data in the analyzed source code could potentially influence the agent's plan generation.
Ingestion points: The agent uses search, sem_search, and read tools to research the project's files as instructed in SKILL.md.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to distinguish between codebase content to be analyzed and instructions to be followed.
Capability inventory: The skill can create and write Markdown files to the plans/ directory and execute shell scripts to validate those plans.
Sanitization: Content retrieved from the codebase is not sanitized or filtered for instruction-like patterns before being processed by the agent.
[COMMAND_EXECUTION]: The skill includes shell scripts (validate-plan.sh and validate-all-plans.sh) that the agent is instructed to run to verify the structural integrity and quality of generated plans.
Execution method: Scripts are called directly via the shell tool to process the newly created plan files.
Security posture: The scripts utilize standard Unix utilities like grep, sed, and cat with appropriate quoting on variables to prevent command injection from file paths. The use of chmod +x in 'validate-all-plans.sh' is limited to the skill's own internal components, which is standard for local tooling.

create-plan