The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands including gofmt, pnpm run format, golangci-lint, and pnpm run lint as part of its quality gate process. It also executes test suites for the code and tests it generates locally.\n- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. It ingests data from external project files (specifications, requirements, and tasks) and passes this data to sub-agents that possess code-writing and command-execution capabilities. There are no boundary markers or sanitization steps to prevent instructions within these files from overriding agent behavior.\n
Ingestion points: Found in references/multi-task-mode.md and references/single-task-mode.md targeting prd/specs/*/tasks.md, spec.md, requirements.md, and **/README.md.\n
Boundary markers: Absent in the sub-agent prompts; instruction blocks are defined but do not contain warnings to ignore embedded content.\n
Capability inventory: Sub-agents launched via the Task tool are tasked with writing code, creating/modifying files, and executing shell commands (linting, formatting, and running tests).\n
Sanitization: No input validation, escaping, or instruction filtering is applied to the content of the ingested project files before processing.

dev-workflow