The Agent Skills Directory

[COMMAND_EXECUTION]: The skill frequently utilizes shell commands for environment discovery and task execution, such as listing architectural decision records (ls plans/adr/ADR-*.md) and running project-specific quality scripts (./scripts/code-quality.sh).
[DYNAMIC_EXECUTION]: The system includes 'meta-agents' like agent-creator and skill-creator designed to generate new task agents or instruction sets (skills) at runtime. This involves writing new files (Write, Edit tools) that are subsequently invoked to perform tasks, representing a script-generation-and-execution pattern.
[INDIRECT_PROMPT_INJECTION]: The skill's core methodology relies on ingesting external data to inform planning decisions.
Ingestion points: Reads architectural decision records from plans/adr/*.md and explores the codebase using the Explore agent.
Boundary markers: The instructions do not define clear delimiters or warnings to ignore embedded instructions within the ingested ADRs or code comments.
Capability inventory: The agents involved have access to high-privilege tools including Bash, Write, Edit, and the ability to spawn further Task or Skill processes.
Sanitization: There is no evidence of validation or sanitization of the content retrieved from project files before it influences the agent's task decomposition and strategy selection.
[DATA_EXFILTRATION]: The skill references tools for external synchronization and research, such as storage-sync (for Turso/redb) and web-search-researcher. While aligned with the stated coordination purpose, these tools facilitate data movement to external services.

goap-agent