The Agent Skills Directory

[SAFE]: The skill implements a structured project management workflow using internal files and delegation to sub-agents. It performs routine file reading and task orchestration without malicious patterns.
[REMOTE_CODE_EXECUTION]: While the skill spawns other agents via the 'Task' tool, these are internal agent-to-agent delegations used for specialized programming tasks (UI, gameplay, AI) and do not involve executing untrusted remote code or piped shell scripts.
[COMMAND_EXECUTION]: The skill uses the 'Bash' and 'Task' tools for project management actions such as running local tests and delegating implementation phases. These operations are scoped to the local development environment and follow standard developer workflows.
[EXTERNAL_DOWNLOADS]: No external network operations, downloads, or remote dependencies were identified in the skill instructions or workflow.
[PROMPT_INJECTION]: The skill contains instructional language for orchestrating other agents, but does not include patterns designed to override system safety guidelines or leak system prompts.
[DATA_EXFILTRATION]: The skill reads internal project documentation (ADRs, GDDs, manifests) to provide context for development, but there are no mechanisms to transmit this data to external or unauthorized endpoints.
[OBFUSCATION]: All instructions and code patterns are in plain text. No encoded strings, homoglyphs, or hidden characters were detected.
[PRIVILEGE_ESCALATION]: The skill does not attempt to use 'sudo', modify system-level directories, or bypass security permissions.
[PERSISTENCE]: The skill maintains 'session state' by writing to a local markdown file ('active.md'), which is a benign tracking mechanism for the development workflow and does not involve shell profile or cron job modification.

dev-story