The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external content (blog posts, talk abstracts, etc.) provided by the user, creating a surface for indirect prompt injection where malicious instructions embedded in the content could potentially override the agent's behavior.
Ingestion points: User-provided content piece (Q&A step 1) and the local configuration file ~/.devadvokit.md.
Boundary markers: Absent. The skill does not employ delimiters or specific instructions to ignore embedded commands within the analyzed material.
Capability inventory: The skill instructions involve reading local files (~/.devadvokit.md and ../../shared/ai-antipatterns.md).
Sanitization: Absent. There is no evidence of logic to filter or escape potential instructional patterns within the ingested content.
[DATA_EXFILTRATION]: The skill accesses a hidden configuration file located at ~/.devadvokit.md. While this is consistent with the tool's stated purpose of utilizing 'DevRel context', accessing hidden files in the user's home directory is a pattern associated with gathering local environment data.
[DATA_EXFILTRATION]: The skill performs directory traversal to read ../../shared/ai-antipatterns.md, accessing a file located outside the immediate skill directory structure. The agent is further instructed to perform this check and any subsequent content rewriting 'silently', without informing the user.

find-content-angles