The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input in the form of code diffs, design documents, and project rules.
Ingestion points: SKILL.md Step 1 (rule files), Step 2 (git diff output), and Step 4 (design documents).
Boundary markers: No explicit delimiters or instructions are used to isolate the data being reviewed from the agent's internal instructions.
Capability inventory: The skill performs file reads, executes git commands, and writes review findings to the filesystem as described in SKILL.md Step 5.
Sanitization: There is no evidence of sanitization or filtering of the content extracted from external files before it is processed by the agent.
[COMMAND_EXECUTION]: The skill relies on executing shell commands to function.
Evidence: SKILL.md Step 2 and references/review-guide.md describe the use of git diff and git log to acquire code changes based on user-supplied parameters such as --base-commit or --task IDs.

spec-review