The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/with_server.py uses subprocess.Popen with shell=True to execute command strings passed as the --server argument. This allows for arbitrary shell command execution if the input is not strictly controlled by the user.
[DYNAMIC_EXECUTION]: Several shell scripts, including scripts/qa-check.sh and scripts/coverage-report.sh, parse repository configuration files (like package.json, Makefile, and Gemfile) and execute the resulting strings using bash -lc "$command". This pattern is used to run project-specific linting, typing, and testing tools dynamically.
[EXTERNAL_DOWNLOADS]: The SKILL.md file defines a mechanism for installing external skills using scripts/install-external-skills.py (e.g., python3 scripts/install-external-skills.py --skill deslop). This allows the agent to fetch and install third-party code at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect execution surface by ingesting and running commands defined in untrusted repository files.
Ingestion points: Reads command strings from package.json, Makefile, pytest.ini, pyproject.toml, and Gemfile across scripts like qa-check.sh and coverage-report.sh.
Boundary markers: None observed; the scripts directly evaluate the content extracted from these files.
Capability inventory: Uses bash -lc and subprocess.Popen(shell=True) to execute the extracted commands.
Sanitization: No validation or sanitization is performed on the command strings before execution.

testing