What problem does it solve? Guidance files such as skills, subagents, rulebooks, partials, and AGENTS.md accumulate filler, redundant rules, and stale rationale over time, inflating the context every agent session loads. Manually trimming them risks deleting load-bearing instructions, and there is no record of which cuts were already considered and rejected. ## Core Features & Use Cases - Graduated cut levels: Choose cautious, conservative, moderate, or aggressive trimming, where each level demands a different amount of evidence before a cut is allowed. - Evidence-based candidate checking: A bundled helper resolves target and transitive files, checks candidates against test assertions and git history, and tracks deployed byte sizes so savings are measured against what agents actually load. - Decline tracking and review markers: Declined cuts are recorded in .agents/streamline-guidance.yaml so they are never re-proposed, and every reviewed file is marked to baseline future growth reports. - Use Case: Point the skill at a bloated SKILL.md and a shared partial, review a numbered table of proposed cuts with safety rationale, apply the approved ones, bump affected rulebook versions, run the quality gate, and commit with a summary of deployed bytes saved. ## Quick Start Use the streamline-guidance skill on my AGENTS.md file at the moderate level and show me the proposed cuts before changing anything.