The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to discover and execute shell commands defined within the target repository (e.g., Makefile, package.json, Cargo.toml). This is an essential component of TDD for running unit tests and full-suite verifications. The instructions correctly emphasize using local repository wrappers (like ./gradlew or ./mvnw) over global tools to ensure environment consistency.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it ingests untrusted data from external bug reports and repository code while having the capability to execute commands and write files.
Ingestion points: The fix-bug workflow ingests user-provided bug reports and reported inputs. The RED step requires reading terminal output from test failures.
Boundary markers: The skill uses strict output formats for its own reporting but does not explicitly define delimiters for raw input data.
Capability inventory: The skill is authorized to perform shell command execution (to run tests) and file modifications (to implement fixes/features).
Sanitization: The instructions provide a strong safeguard by requiring the agent to "Read the failure, not just the red" (Rule 5) and verify that the test fails specifically because the behavior is missing, which helps mitigate accidental or malicious execution of instructions embedded in data.

test-driven-development