The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to execute multiple local maintenance and registration scripts using node and domain-specific CLI tools.
Evidence: node scripts/verify-blind-rubric.mjs, node scripts/register-cheat-score-evidence.mjs, node scripts/stamp-episode-script-hash.mjs, node scripts/prepare-cheat-prediction-input.mjs, node scripts/register-cheat-prediction.mjs, and laohan-bianpai vendors --sync.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied content (video scripts and outlines) which influences automated scoring and prediction workflows, creating a vulnerability surface for instruction injection.
Ingestion points: Reads user-provided files from the episodes/<slug>/ directory, specifically 01-口播稿.md and 00-选题.md.
Boundary markers: The instructions do not define clear delimiters or warning markers to prevent the agent from obeying instructions embedded within the processed video scripts.
Capability inventory: The skill has the capability to execute shell commands via node and laohan-bianpai, and it performs file write operations to the calibration/ and predictions/ directories based on the processed content.
Sanitization: No evidence of input validation, sanitization, or filtering is present for the content ingested from the markdown scripts before they are passed to the scoring and registration scripts.

laohan-jiaozhun