The Agent Skills Directory

[PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection by processing unstructured, potentially untrusted input to influence its decision-making logic and command generation.
Ingestion points: External data enters through fields such as decision_needed, blocking_items, and conflict_description as described in the Invocation Pattern section of SKILL.md.
Boundary markers: The prompt templates do not include explicit delimiters or instructions to ignore embedded commands within these descriptive fields.
Capability inventory: The skill generates orders (e.g., "Split PR", "Require ADR") that are intended for consumption by calling commands like /do-issue and /fix-pr, which possess repository management and file system capabilities.
Sanitization: There is no mention of sanitization or validation for the input data before it is processed by the reversibility assessment (RS) logic.
[COMMAND_EXECUTION]: The documentation includes a verification step for developers that involves the execution of the make test-checkpoint shell command.

war-room-checkpoint