The Agent Skills Directory

[PROMPT_INJECTION]: The skill creates an indirect prompt injection surface by converting user-supplied PRD content into tasks that dictate the agent's work plan.
Ingestion points: Reads feature PRDs and task lists from the .plan/ directory.
Boundary markers: Uses Markdown headings for data organization without dedicated safety delimiters for untrusted input.
Capability inventory: Can write files, move directories, and invoke other system skills.
Sanitization: PRD content is processed without prior validation or escaping.
[COMMAND_EXECUTION]: The skill uses shell commands during installation to set up the project planning structure and copy its internal command definitions to the project's local command directory. These operations are project-local and align with the described setup process.
[SAFE]: Analysis of the skill instructions and references shows no evidence of obfuscation, data exfiltration, or malicious persistence. The operations performed are consistent with a standard project management utility.

launchpad