The Agent Skills Directory

[SAFE]: The skill utilizes standard utilities such as git show, wc -l, and shell timers (time, Measure-Command) to perform quantitative analysis of project files. These operations are restricted to the local environment and are necessary for the skill's primary function.
[SAFE]: No network access or data exfiltration patterns were identified. Baseline measurements are stored in a dedicated local memory-tier directory and are explicitly marked as machine-bound and never to be committed.
[SAFE]: The instructions include robust safeguards against deceptive behavior, requiring the agent to report honestly when a baseline is unavailable rather than fabricating improvement deltas.
[SAFE]: The skill demonstrates a least-privilege approach to tool usage by only invoking specific marketplace plugin skills (like dotnet-diag or cloudflare:web-perf) if they are already installed and applicable to the project stack.
[SAFE]: While the skill ingests project source code to calculate metrics (creating an indirect prompt injection surface), the risk is negligible given its focus on quantitative measurement and lack of high-risk autonomous actions based on code content.
Ingestion points: Project source code files (via git show) and tool outputs.
Boundary markers: Absent.
Capability inventory: Local file reading, execution of counting/timing utilities, and invocation of other verification skills.
Sanitization: None.

measure