The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill accesses workspace transcripts and personal skill directories to identify user patterns.
The skill reads chat logs from the workspace-scoped agent-transcripts/ directory.
The skill recursively searches for existing skill files in ~/.cursor/skills/ and project-level .cursor/skills/ directories.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted historical data from chat transcripts, which presents a surface for indirect prompt injection attacks where historical content could influence the generated skill output.
Ingestion points: Workspace chat transcripts located in the agent-transcripts/ directory.
Boundary markers: No explicit delimiters or boundary markers are defined to separate historical data from the instructions provided to the mining subagents.
Capability inventory: The skill utilizes create-skill for file authoring, orchestrates subagents, and performs Git operations including worktree creation, commits, and pull requests.
Sanitization: The instructions do not specify any sanitization, filtering, or validation of the historical transcript content before it is processed by the analysis subagents.

automate-me