The Agent Skills Directory

[COMMAND_EXECUTION]: The validator script scripts/check.py uses subprocess.run to execute internal Python tools (validate_skill.py, validate_evals.py) and the reporter script. These commands are constructed as lists using the system's Python executable and deterministic local file paths, preventing shell injection.
[PROMPT_INJECTION]: The skill implements logic to analyze the evals/evals.json files of other skills in the repository. This represents a potential indirect prompt injection surface.
Ingestion points: scripts/trace_report.py reads and parses evals/evals.json from every subdirectory within the repository's skills/ folder.
Boundary markers: No explicit boundary markers or instructions to ignore embedded commands are present when reading the JSON content, as the data is used for static regex matching rather than LLM interpolation.
Capability inventory: The skill has the capability to execute shell commands (via scripts/check.py) and read files throughout the skill tree.
Sanitization: The prompts read from external evals.json files are processed via regular expressions to detect invocation patterns but are not otherwise sanitized or escaped.

skill-trace-debugger