The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill generates a shell script (expected.sh) at runtime and executes it via bash to process test cases. It also executes agent-generated model answers (e.g., using python3) to produce expected output files, which involves running code derived from user input.
[COMMAND_EXECUTION]: The skill instructs the agent to execute various shell commands, including bash, python3, bunx, and mise, to automate the problem development and validation lifecycle.
[EXTERNAL_DOWNLOADS]: The skill uses bunx (npm) to execute the @exercode/problem-utils package and uses mise install to download and set up runtimes and command-line tools based on the repository's configuration files.
[PRIVILEGE_ESCALATION]: The instructions include the command mise trust --yes, which bypasses the manual security prompt for trusting configuration files (mise.toml). This allows the tool to execute potentially unverified configuration tasks without user intervention.
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it ingests user-provided source material to generate code that is subsequently executed.
Ingestion points: SKILL.md Step 1 (user-provided learning objectives and source material).
Boundary markers: Absent.
Capability inventory: python3 (Step 6), bash (Step 6), and bunx (Steps 7, 8, 10).
Sanitization: Absent (relies on agent reasoning and downstream validation tools).

generate-judge-problems