The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external design documents and repository instructions to determine implementation scope and manage GitHub issues. This creates a surface for indirect prompt injection where malicious instructions embedded in design content could influence the agent's actions during the delivery workflow.\n
Ingestion points: Workflow Step 1 in SKILL.md requires reading repository instructions and the complete design document.\n
Boundary markers: No specific delimiters or safety instructions are defined to separate the design content from the agent's core instructions.\n
Capability inventory: The skill orchestrates file writes, GitHub PR/Issue management (via sub-specialists), and local command execution as seen in tests/run_all.sh.\n
Sanitization: The instructions do not specify any validation or sanitization for the content of the design documents before they are used to drive the workflow.\n- [COMMAND_EXECUTION]: The tests/run_all.sh script executes shell and Python scripts using relative paths within the repository structure. This involves running external script files during the testing phase.\n
Evidence: The script runs bash -n on shell scripts and python3 on test files like story_contract_test.py and render_issue_body_test.py located in sibling skill directories.

feature-delivery