The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external project files to define its testing scope and parameters, creating a surface for indirect prompt injection where malicious content in those files could influence the agent's actions during the adversarial loop.\n
Ingestion points: The skill reads .cheese/cook/<slug>.md for handoff instructions and .cheese/glossary/<slug>.md for canonical terms.\n
Boundary markers: There are no explicit delimiters or instructions provided to the agent to ignore potentially malicious embedded content within the ingested files.\n
Capability inventory: The skill can execute shell commands via python3, write new files to the .cheese/press/ directory, and execute dynamically generated test code.\n
Sanitization: The instructions do not specify any sanitization, validation, or escaping of the content read from the external files before it is processed.\n- [DYNAMIC_EXECUTION]: The skill is designed to dynamically generate and execute adversarial tests, fixtures, and test-only harness support at runtime based on the approved contract and handoff data.\n
Evidence: The 'Adversarial loop' section explicitly instructs the agent to 'Add or run only tests, fixtures, and test-only harness support' and to 'Replay the same attack'.\n- [COMMAND_EXECUTION]: The skill utilizes a local Python zip application (press.pyz) to perform administrative tasks, resolving slugs and writing artifacts.\n
Evidence: Multiple instructions call for the execution of python3 skills/press/scripts/press.pyz with various subcommands like wheypoint-resolve and write-handoff-artifact.

press