The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill includes scripts that use Playwright to browse external URLs and local files to measure layout geometry and extract text content. This creates a surface for indirect prompt injection if the processed content contains malicious instructions. However, the risk is mitigated by the skill's focus on geometric data, text truncation (80 characters), and strict evidence-based reasoning rules.
Ingestion points: scripts/measure-layout.mjs, scripts/detect-overflow.mjs, and scripts/detect-collisions.mjs collect data from URLs and files via the layout-probe.mjs shared utility.
Boundary markers: The experience-core.md file defines strict operational rules, including a 'Verdict before adjectives' requirement and evidence-level labeling (E1-E6) to prevent the agent from blindly following content.
Capability inventory: The skill executes local Node.js scripts and uses Playwright to perform browser-based measurements. It does not have write access to the user's environment beyond the provided scripts.
Sanitization: The data collection script (collectSnapshot) truncates extracted text to 80 characters, limiting the payload of potential injections.

composition-repair