The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs various administrative actions using the coval CLI, including metrics update and review-annotations update. These operations are central to the skill's purpose and are protected by a requirement for explicit user confirmation before execution.
[DATA_EXFILTRATION]: The skill retrieves the COVAL_API_KEY from the local CLI configuration or environment variables to authenticate requests to the vendor's API. This is a legitimate use of credentials for authorized data retrieval from the official platform.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to api.coval.dev using curl to fetch extended metric information. This targets the vendor's official infrastructure and is consistent with the skill's operational requirements.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes untrusted data from simulation transcripts and reviewer notes.
Ingestion points: Transcript arrays and reviewer_notes fields are fetched in Phase 1, Step 2.
Boundary markers: No specific delimiters or instructions are used to isolate these inputs from the analysis prompt.
Capability inventory: The skill can execute shell commands to update metric prompts (coval metrics update) and modify human labels (coval review-annotations update).
Sanitization: No sanitization or filtering of the transcript or notes content is described before they are analyzed by the LLM.

review-llm-annotations-and-improve-prompt