The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted external data, specifically code diffs and requirement specifications, which are used to map behaviors and generate tests. This creates a surface for indirect prompt injection where malicious instructions embedded in the code or specs could influence the agent's actions.
Ingestion points: Reads specs, acceptance criteria, and 'cooked diffs' in SKILL.md.
Boundary markers: Absent; there are no explicit delimiters or instructions to ignore embedded commands within the processed data.
Capability inventory: Uses cheez-write for file modification and shell commands for executing tests.
Sanitization: No sanitization or input validation is performed on the ingested data before use.
[COMMAND_EXECUTION]: The skill dynamically generates test code and executes shell commands to run these tests.
Evidence: SKILL.md describes a workflow (Steps 4 and 6) where tests are written via cheez-write and then executed via the project's gates/commands. This represents the execution of dynamically created content.

press