The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to handle fuzzer corpus and crash files which are untrusted by nature. While the included inventory-corpus.py script safely catalogs these files using hashing, the fuzzer itself is intended to execute code based on these inputs. This represents an attack surface for indirect injection if malicious instructions were embedded in a crash reproducer.
Ingestion points: Fuzzer corpus and artifact directories processed in scripts/inventory-corpus.py and by the Jazzer tool.
Boundary markers: The skill explicitly instructs users to isolate the fuzzing environment, avoid using live secrets, and perform testing within authorized workspaces.
Capability inventory: Execution of build tools (mvn, gradle, bazel) and the Jazzer binary launcher; file system read access for the Python inventory script.
Sanitization: The inventory-corpus.py script processes files as binary streams to compute SHA-256 hashes without interpreting the content.
[COMMAND_EXECUTION]: The skill directs the agent to execute standard development and testing commands, such as Java compilation (javac), build lifecycle tasks (mvn test), and the execution of the Jazzer native launcher binary. These operations are standard for the tool's intended use.
[EXTERNAL_DOWNLOADS]: The instructions guide the user to download Jazzer release binaries and dependencies from official project repositories. The skill correctly suggests integrity verification using SHA-256 checksums to mitigate supply chain risks.

jazzer