The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Risk Surface. The skill is designed to ingest and edit large research reports, creating a surface where malicious instructions embedded in documents could influence the agent to perform unauthorized filesystem operations.
Ingestion points: Document files are read from the local filesystem during the editing workflow.
Boundary markers: Prompt templates lack explicit delimiters to isolate the document content from agent instructions.
Capability inventory: The skill possesses file read and write capabilities (surgical patches).
Sanitization: No input validation or sanitization of document content is described.
[PROMPT_INJECTION]: Self-Referential Analysis. The inclusion of skill-report.json with a pre-authored security audit that claims the skill is safe and dismisses potential flags is a self-referential pattern used to interfere with automated security evaluations.
[PROMPT_INJECTION]: Metadata Inconsistency. There is a mismatch between the author stated in the skill's metadata ('ACSKamloops') and the author provided in the analysis context ('aiskillstore'), which can be used to mask the skill's true source.

large-report-editor