The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes data from external APIs (theses.fr and HAL), such as thesis titles and abstracts, which are not under the skill's direct control.
Ingestion points: API responses from https://theses.fr/api/v1/ and api.archives-ouvertes.fr/search/tel/.
Boundary markers: None identified in the prompt templates to separate external data from instructions.
Capability inventory: The skill utilizes Python (urllib.request) for network operations and provides instructions for shell command execution.
Sanitization: No specific escaping or sanitization logic is described for the incoming text data beyond JSON structure normalization.
[COMMAND_EXECUTION]: The 'Procédure de récupération en cas de breakage' section provides specific shell commands (curl and grep) for the agent to execute to diagnose API changes and discover new endpoints from the theses.fr website.
[EXTERNAL_DOWNLOADS]: The skill connects to well-known academic and governmental domains including theses.fr, archives-ouvertes.fr, idref.fr, and sudoc.fr to retrieve research data and PDF documents.

recherche-theses-allison-fiorentino