lint

Process linting detects stalled behavior and misconfigurations by scanning agents, skills, and hooks against a ruleset. It matches executed work against known problem patterns to identify issues and produces lint results with counts and evidence before fixes are attempted.

Ruleset-driven process linting evaluates automation components against curated global rules to prevent recurring failure modes. It optionally augments checks with domain-specific rules, applying tiered responses that automatically fix low-risk patterns, warn with evidence, or record observations for watch rules.

To validate agent configurations, use the improve:lint trigger to scan a specific process. The tool loads global lint rules from the references directory, optionally invokes domain improve skills for additional rules, and evaluates detection steps per applicable rule tier to surface misconfigurations.

Yes, tiered responses enable safe autonomy by automatically applying fixes for low-risk patterns. For higher-risk issues, the process linting surfaces evidence for warnings and records observations for watch rules without making changes itself.

Ruleset management allows you to add new rules from observations, promote or demote tiers based on evidence, and review the current ruleset. This helps maintain configuration validation standards without requiring you to make fixes directly during the review.

Agents drift into recurring failure modes when processes are not checked against known problem patterns. Process linting catches these issues by matching executed work against a curated ruleset of misconfigurations and stale references before they cost tokens.