The Agent Skills Directory

[NO_CODE]: The skill consists entirely of Markdown documentation and a JSON metadata report. No functional code, shell scripts, or executable binaries are included in the package.
[METADATA_POISONING]: The skill-report.json file contains deceptive and misleading metadata. Specifically, the provenance section references a non-existent AI model (gpt-5.5) and provides a source URL (github.com/CodingCossack/...) that contradicts the platform's attribution of the skill to the author aiskillstore. This metadata appears designed to simulate a legitimate audit trail through hallucinated authorities.
[INDIRECT_PROMPT_INJECTION]: The skill defines a workflow surface for indirect prompt injection within the 'Agent Prompt Template' section of SKILL.md. The instructions encourage agents to ingest untrusted external data without sufficient sanitization or boundaries. • Ingestion points: Subagents are instructed to read [test name] and [error summary] from external test logs/files (SKILL.md lines 65-81). • Boundary markers: The provided prompt template lacks delimiters or instructions for the subagent to ignore potentially malicious content embedded in the test failures. • Capability inventory: The workflow grants subagents the capability to modify the local file system to 'Fix' code and return a summary of changes. • Sanitization: No escaping, validation, or filtering of the external test output is mentioned in the workflow instructions.

dispatching-parallel-agents