The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses directives to control agent behavior (e.g., preventing sub-task delegation) and established personas (Lead, Architect, Skeptic), which are functional for the orchestration task and do not target safety filter bypass or system prompt extraction.
[DATA_EXFILTRATION]: The instructions include a non-negotiable principle (Shared Principle #4) that explicitly prohibits secrets, API keys, and PII from appearing in prompts, messages, or artifacts, demonstrating a high awareness of data protection.
[COMMAND_EXECUTION]: The orchestration relies on platform-native tools like TeamCreate and Agent to manage sub-processes. There are no instances of arbitrary shell command execution or privileged system modifications.
[EXTERNAL_DOWNLOADS]: All file access is restricted to the local project directories (docs/, plugins/, etc.). No external scripts or remote dependencies are downloaded or executed during the skill's operation.
[SAFE]: The skill incorporates a structured 'Skeptic Protocol' and 'Contract Negotiation' phase, ensuring that all proposed changes are verified against the original technical specification and quality standards before finalization.

plan-implementation