The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a Python helper script (scripts/profile_pipeline.py) to execute the glab CLI for GitLab API interactions. These commands are executed as lists using subprocess.run(), preventing shell injection. The script also runs itself under python3 for various analysis stages.
[INDIRECT_PROMPT_INJECTION]: The skill processes job trace logs fetched from an external source (GitLab CI), which introduces a surface for indirect prompt injection if an attacker can control job output to manipulate the agent's interpretation.
Ingestion points: Trace logs are fetched via glab api in the cmd_trace function and then read by the sections and signals commands in scripts/profile_pipeline.py.
Boundary markers: The skill relies on a predefined flow and report template but does not explicitly implement delimiters or safety warnings to ignore instructions embedded within the processed logs.
Capability inventory: The agent has the capability to read local project files, including .gitlab-ci.yml, Makefile, and Dockerfile. It produces a markdown report but does not have the capability to write to the filesystem or execute arbitrary external code based on log content.
Sanitization: The signals command uses specific regular expressions to extract metrics (e.g., go: downloading, sleep, buildx steps), which serves as a structured filter that reduces the likelihood of raw malicious text being interpreted as instructions by the model.

ci-pipeline-profiler