The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs several local command-line operations to manage git branches and create pull requests.
It uses git merge-base, git checkout, git pull, and git merge to integrate code.
It uses the GitHub CLI (gh pr create) to automate pull request creation.
It uses a custom script scripts/run-tests to execute the project's test suite.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through project configuration files that it parses and acts upon.
Ingestion points: The scripts/run-tests script reads content from package.json, Cargo.toml, pytest.ini, conftest.py, setup.cfg, go.mod, and Makefile within the local project directory.
Boundary markers: None. The script does not use delimiters or warnings when processing these files.
Capability inventory: The skill has the capability to execute shell commands (exec $RUNNER), write to the terminal, and perform git operations (push, merge, delete branches).
Sanitization: There is no sanitization or validation of the commands extracted from configuration files (specifically the test script in package.json or targets in Makefile) before they are passed to the shell for execution.
[DYNAMIC_EXECUTION]: The scripts/run-tests utility implements dynamic execution by auto-detecting a project's test runner and executing it.
The script uses sed to extract command strings from package.json and then uses exec $RUNNER to run the result.
This allows for the execution of arbitrary commands defined in the project's metadata, which is a standard feature for development tools but presents a security risk if the agent is operating on an untrusted or malicious repository.

finishing-a-development-branch