The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches and processes untrusted data from external Author Guidelines in Phase 2: Data Extraction to generate journal profiles. The external content is not isolated with boundary markers or specific instructions to ignore embedded prompts, which could allow malicious instructions in the guidelines to influence the agent's behavior. This risk is mitigated by manual review checkpoints in Gate 1 and Gate 2.
Ingestion points: SKILL.md Phase 2 (Fetching an Author Guidelines URL or processing user-pasted text).
Boundary markers: Absent; external data is parsed directly without delimiters or safety warnings to disregard instructions within the data.
Capability inventory: The skill uses Write and Edit tools to create and modify profiles in Phase 4, and executes a local validation script.
Sanitization: Absent; the guidelines content is processed without filtering or escaping.
[COMMAND_EXECUTION]: The skill.yml manifest defines a validation_commands entry that executes a local script: python3 scripts/validate_catalog_consistency.py. This script is used to ensure the profile counts remain synchronized during public contributions.

add-journal