Eagle Skill Optimizer

Problem

Many SKILL.md files do not fail outright after deployment; they drift. Triggering may be unstable, key steps may be skipped, earlier context may be lost on large files, output structure may change run by run, or users may be unsure whether a write to a production system actually took effect. Optimizing by feel can turn one fragile constraint into another. eagle-skill-optimizer turns skill maintenance into a traceable engineering review: confirm the target file and symptoms, diagnose it against 18 engineering patterns, and apply only the changes the user selects.

How It Works

• Diagnose: Read the target SKILL.md and classify scenarios such as read-only, no external dependencies, or short-flow skills, so non-applicable patterns are not reported as missing. Checks cover workflow stability, upfront information collection, context retention, evidence-driven output, output templates, progress visibility, draft confirmation, and write receipts.
• Report: Return prioritized high / medium / low recommendations. Each item is tied to a specific pattern and explains the gap, why it matters, and the concrete fix.
• Modify: Back up the original file before changes. Drafts are shown in conversation first; files are written only after explicit confirmation. If the user cancels, restore from backup and provide a completion receipt.

Boundaries

It is best used to audit and patch the stability of existing agent skills, not to invent business rules from scratch. In read-only Q&A, short-flow, or no-external-tool scenarios, some patterns are marked not applicable. When a skill writes to production systems, calls external services, or performs irreversible operations, manually confirm data boundaries, permissions, and rollback policy.