The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill explicitly instructs the agent to use sudo to install runner tools on Linux systems.
Evidence: curl -L https://packages.gitlab.com/.../script.deb.sh | sudo bash and sudo apt-get install gitlab-runner.
Evidence: curl -L https://raw.githubusercontent.com/nektos/act/master/install.sh | sudo bash.
[CREDENTIALS_UNSAFE]: The skill defines a convention for storing and accessing sensitive CI/CD secrets in a specific local file path.
Evidence: Accessing ~/.lenneTech/ci-secrets.env to source variables like $CI_TOKEN or $DEPLOY_KEY.
[EXTERNAL_DOWNLOADS]: The skill fetches installation scripts from external sources for toolchain setup.
Evidence: Downloads configuration and scripts from packages.gitlab.com (GitLab) and raw.githubusercontent.com/nektos/act.
[DYNAMIC_EXECUTION]: The skill dynamically generates executable configuration files and shell commands at runtime based on the contents of the repository.
Evidence: Step 5 Path B involves generating a docker-compose.ci.yml file on the fly and executing it with docker compose.
Evidence: The command field in the generated YAML is constructed by joining strings from untrusted CI configuration files.
[INDIRECT_PROMPT_INJECTION]: The skill processes and executes instructions from untrusted data sources within the repository.
Ingestion points: Reads .gitlab-ci.yml and .github/workflows/*.yml from the project directory.
Boundary markers: None. The skill directly executes content found in the script or command fields of these files.
Capability inventory: Full shell execution via gitlab-runner, act, and docker compose, as well as file system read/write access.
Sanitization: None. The skill assumes the content of the CI configuration files is safe to execute in a shell environment.

validating-ci-pipelines-locally