The Agent Skills Directory

[SAFE]: The skill executes a legitimate development workflow for finalizing code changes and documentation within a controlled repository environment.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute standard developer commands like git and gh (GitHub CLI), and runs local Python scripts via uv run from the relative project path ../dstack-core/scripts/. These operations are appropriate for its stated purpose of feature delivery.
[PROMPT_INJECTION]: The skill processes repository files and user-supplied strings which are used in subsequent workflow steps. It manages the risk of indirect prompt injection by incorporating mandatory evidence chains and human oversight: 1. Ingestion points: The skill reads feature documentation (design.md, SUMMARY.md) and accepts user-provided feature selectors. 2. Boundary markers: Instructions explicitly mandate that the agent must "Inspect commits, implementation, tests, and changed files before deciding whether documentation is accurate" and use the AskUserQuestion tool for delivery decisions. 3. Capability inventory: The skill has access to Bash, Write, and Edit tools to perform filesystem mutations and command execution. 4. Sanitization: While not using automated filtering, the workflow requires human verification of all states and audit rows before closing a feature.

close-feature