The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill uses user-provided requirements to construct queries for a BM25 retrieval engine and eventually assembles them into shell commands, creating an attack surface for indirect injection.
Ingestion points: Ingests untrusted data through the interactive requirement gathering step in SKILL.md and references/requirement-extraction.md.
Boundary markers: Lacks explicit delimiters or instructions to ignore embedded commands in user queries.
Capability inventory: Generates gst-launch-1.0 commands and executes validation via subprocess.run in scripts/validate_pipeline.py.
Sanitization: The validator script implements a regex-based allowlist for element names and uses shlex.split for command tokenization to prevent shell injection.
[DYNAMIC_EXECUTION]: The scripts/validate_pipeline.py script executes gst-launch-1.0 and gst-inspect-1.0 via subprocess.run to verify the syntax and structural integrity of generated pipelines. This involves executing logic based on dynamically assembled content derived from the skill's dataset and user input.
[COMMAND_EXECUTION]: The primary output of the skill is a runnable shell command. The skill instructions in SKILL.md (Step 5) and references/output-format.md explicitly forbid dangerous shell patterns such as heredocs, variable indirection, and line continuations to minimize risks.

deepstream-generate-pipeline