The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted external data, including paper titles, abstracts, and community signals from academic databases and social platforms, which could contain malicious prompts meant to influence the agent's evaluation logic.
Ingestion points: Data enters the skill's context via the research-literature-search sub-skill and supplementary signals from URLs (e.g., Hugging Face, X, and researcher blogs) as specified in SKILL.md.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to treat ingested paper metadata as non-executable text or to ignore embedded instructions.
Capability inventory: The skill possesses the ability to write to the local filesystem (creating directories in papers/ and .bensz-api/) and execute shell commands (specifically gh api) as described in SKILL.md and performed by scripts like scripts/reconcile_integrity.py.
Sanitization: While scripts/catalog.py performs title normalization for ID generation, the full metadata content is archived without significant sanitization.
[COMMAND_EXECUTION]: The skill instructions in SKILL.md authorize the use of the gh api command-line tool for uploading bug reports.
Evidence: The skill specifies that bug reports are first recorded to ~/.bensz-skills/bugs/ and only uploaded to GitHub using gh api if the user explicitly requests it. This represents a controlled interaction with a well-known service (GitHub) for diagnostic purposes.

research-literature-radar