The Agent Skills Directory

[SAFE]: No malicious patterns, obfuscation, or security vulnerabilities were detected in the skill's instructions, metadata, or scripts.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from academic papers (papers/fulltext/*.txt) to generate structured records. This represents a potential ingestion surface for indirect prompt injection. However, the skill implements specific sanitization logic in scripts/run.py using regex patterns from assets/source_text_hygiene.json to filter out promotional and narrative author prose. The output is constrained to a machine-readable JSONL format, which reduces the likelihood of an agent misinterpreting embedded data as instructions.
[COMMAND_EXECUTION]: The skill uses a local Python script (scripts/run.py) for deterministic scaffolding and evidence bank construction. The script's operations are limited to local file system reads and writes within the workspace, utilizing standard libraries for CSV and JSON processing. The provided uv run command is a standard development practice for executing local scripts in a controlled environment.

paper-notes