The Agent Skills Directory

[PROMPT_INJECTION]: The skill possess a surface for indirect prompt injection by ingesting untrusted data into generated templates.
Ingestion points: User-provided Acceptance Criteria from the blueprint are interpolated into implementation plans as described in SKILL.md.
Boundary markers: No explicit delimiters or instructions to disregard embedded commands are present in the templates.
Capability inventory: The skill uses Read, Write, and Edit tools to create documentation; it does not autonomously execute code from the generated plans.
Sanitization: No validation or filtering is performed on user-supplied input before interpolation.
[DATA_EXFILTRATION]: The skill manages plan documentation within the local repository's docs/plans/ directory. It does not harvest credentials or exfiltrate data to external domains.
[REMOTE_CODE_EXECUTION]: No patterns for downloading or executing remote code were detected. The skill relies on static templates for plan generation.
[COMMAND_EXECUTION]: No shell commands or privilege escalation patterns are present in the skill files. Mentioned DevOps tools (e.g., majestic-devops:opentofu-coder) are used as procedural references within the templates.

plan-builder