The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill aggregates data from external and untrusted sources, including user reports, issue trackers, and error logs, to reconstruct the user's working context. This creates a vulnerability surface where adversarial instructions embedded in tickets or chat history could influence agent behavior.
Ingestion points: Chat history transcripts, thread lists, local rollout JSONL files, GitHub issue trackers, and error tracking logs.
Boundary markers: The skill relies on citation requirements but lacks explicit structural delimiters or "ignore instructions" warnings for ingested data.
Capability inventory: The skill is authorized to read local thread histories, file rollout paths, and execute git and gh CLI commands.
Sanitization: Includes a high-level instruction to "sanitize private context before any public output," but lacks technical validation or escaping mechanisms for the data it processes.
[COMMAND_EXECUTION]: The skill instructions direct the agent to use system tools, specifically git and gh (GitHub CLI), to verify the live state of pull requests, branches, and tickets.
[EXTERNAL_DOWNLOADS]: The skill is a mirrored version of content from the official Cursor plugins repository on GitHub.

cursor-recall