The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external websites (Author Guidelines) and user-pasted content to generate profiles that are written to the filesystem.
Ingestion points: Phase 2 involves fetching a user-provided URL or accepting pasted guidelines text.
Boundary markers: Features two explicit gates (Gate 1: Metadata Confirmation and Gate 2: Profile Review) where the user must approve the extracted and generated content before it is finalized.
Capability inventory: Employs the Write tool to save profiles to the repository and local $HOME directories.
Sanitization: Relies on human oversight at the confirmation gates to identify and prevent the processing of malicious instructions embedded in the ingested guidelines.
[COMMAND_EXECUTION]: The skill configuration includes a shell command for catalog validation.
Evidence: The skill.yml file specifies python3 scripts/validate_catalog_consistency.py as a validation command.
[DATA_EXFILTRATION]: The skill reads from and writes to paths in the user's home directory and uses relative traversal to access directories outside its own installation.
Evidence: Accesses $HOME/.claude/private-journal-profiles/ for user-local overrides and uses ../ paths to interact with the sibling write-paper and find-journal skill directories.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to arbitrary domains to retrieve information.
Evidence: Phase 2 includes a WebFetch operation to retrieve content from external journal author guidelines URLs provided by the user.

add-journal