The Agent Skills Directory

[COMMAND_EXECUTION]: The skill guides the agent to execute shell commands for running tests and verifying the environment. This is a core functionality of the skill and includes instructions to report exact commands and results to the user.
[DATA_EXFILTRATION]: The skill accesses project-level files such as manifests, configuration, and source code to build a testing context. It includes explicit instructions in resolve-project-testing-context.md to never store secrets, credentials, or absolute machine-specific paths in its persistent context file.
[PROMPT_INJECTION]: The skill processes untrusted repository content (source code, existing tests, manifests) which represents an indirect prompt injection surface.
Ingestion points: Repository instructions, manifests, lockfiles, production code, and existing tests are read from the project root.
Boundary markers: No explicit delimiters for external content are defined, but the skill requires the agent to verify all context facts against current repository evidence before use.
Capability inventory: The agent can execute shell commands (running tests) and perform file operations (creating/updating tests and context files).
Sanitization: The agent is instructed to refresh information if evidence changes and compare context facts with live configuration and build manifests.

ptlam-testing