The Agent Skills Directory

[PROMPT_INJECTION]: Indirect prompt injection vulnerability. The skill accepts user input via the $ARGUMENTS variable and interpolates it directly into the prompt strings used to spawn subagents (e.g., in references/agent-phases.md). There are no boundary markers or instructions for the subagents to ignore instructions contained within the feature description, which could allow a malicious user to hijack the subagents' behavior.\n
Ingestion points: $ARGUMENTS in SKILL.md.\n
Boundary markers: Absent in subagent prompt templates.\n
Capability inventory: Bash, Write, Edit, CronCreate, Agent (High capability).\n
Sanitization: None identified.\n- [COMMAND_EXECUTION]: Extensive use of the Bash tool for project scaffolding, git worktree management (via scripts/worktree-setup.sh), dependency scanning, and test execution across all phases.\n- [EXTERNAL_DOWNLOADS]: The skill performs remote data retrieval using WebFetch for documentation (e.g., docs.example.com) and uses curl or wget within verification scripts to test API endpoints.\n- [DATA_EXFILTRATION]: Employs the PushNotification tool to transmit feature names and test results to an external notification service upon completion of implementation tasks.

implement