The Agent Skills Directory

[SAFE]: No security concerns, malicious patterns, or obfuscation were detected. The skill is entirely instructional.
[NO_CODE]: The skill consists of a single markdown file containing documentation and does not bundle any executable scripts, binaries, or dependencies.
[COMMAND_EXECUTION]: The framework instructions guide the agent to use the Bash tool for local developer tasks such as npm test and grep. These operations are transparent, standard for development, and conducted on local project files.
[INDIRECT_PROMPT_INJECTION]: The skill establishes a workflow for the agent to ingest and process data from project-local evaluation files.
Ingestion points: The agent is instructed to read evaluation definitions and logs from the .claude/evals/ directory.
Boundary markers: The templates utilize Markdown headers and structured lists to organize data, but they do not define explicit delimiters to isolate external instructions from the agent's system prompt.
Capability inventory: The skill uses Bash, Write, Edit, Read, Grep, and Glob tools to perform its tasks.
Sanitization: The framework does not describe or implement automated sanitization or validation for the ingested markdown content.

eval-harness