The Agent Skills Directory

[SAFE]: The skill provides purely instructional guidance for a document review process and does not contain any executable malicious code.
[DATA_EXPOSURE]: The instructions include a robust security invariant: 'MUST NOT send content to an external provider without an explicit human action; redact secrets by default.' This proactively mitigates the risk of accidental credential or data leakage during the review process.
[COMMAND_EXECUTION]: While the skill mentions the potential use of a 'CLI runner' within its advisory workflow, it explicitly instructs the agent to stop and wait for human input rather than executing commands automatically.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external LLP documents. This attack surface is addressed through the following characteristics:
Ingestion points: Workflow step to 'locate and read the document fully'.
Boundary markers: The skill relies on a structured review format (Assessment, Strengths, Concerns, etc.) to compartmentalize the analysis of the untrusted document.
Capability inventory: The skill is authorized to write local artifact files under llp/reviews/.
Sanitization: Explicit instruction to redact secrets by default during processing.
[EXTERNAL_DOWNLOADS]: The skill references its source repository (ccheever/llp@v0.5.2) for context and process guidelines, which is consistent with its stated purpose.

llp-review