The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using the Bash tool based on file paths extracted from external story files. In Phase 3, the agent is directed to "run it via Bash" if a test file path is mentioned in the story's "Test Evidence" section, which constitutes a direct sink for potentially malicious path strings or commands.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from story files found in production/sprints/ or provided via command line, using this data to influence high-privilege actions.
Ingestion points: Story files (markdown), production/sprints/ directory, and user-supplied arguments.
Boundary markers: None; the skill treats the content of the story files as authoritative instructions for verification.
Capability inventory: Bash (command execution), Edit (file writing), and Task (sub-agent spawning).
Sanitization: None; the agent is instructed to run the extracted paths without validation or escaping.
[DYNAMIC_EXECUTION]: The skill dynamically identifies and executes scripts at runtime based on the "Story Type" and extracted metadata, creating a risk if the execution environment is not strictly controlled.
[ADVERSARIAL_ANALYSIS]: The skill includes a --review solo mode that allows the user to skip Phase 5 (Lead Programmer Code Review). This bypasses human-in-the-loop safety gates, allowing the automated Bash execution in Phase 3 to occur and the results to be logged to the session state without a secondary review of the implementation or test code.

story-done