wenyu0608/token-saver
Save AI coding tokens without losing the critical context needed for correct answers.
Review or continue existing code changes using the Git diff, changed symbols, and minimal dependency context. Use for PR reviews, regressions, or work-in-progress changes; do not use for an unfamiliar repository with no relevant diff.
Retrieve only the sections of large local documentation needed to answer a focused question. Use for Markdown, text, API specifications, or documentation trees; do not use when the user requests a complete reading or summary.
Run a noisy build, test, lint, or log command while keeping full output on disk and returning only actionable failures. Use for commands expected to produce large repetitive output; do not use when exact full stdout is the requested artifact.
Preserve and inspect complete in-scope evidence for exhaustive debugging, verification, audits, and high-risk correctness work. Use when omissions could change the conclusion; do not use for ordinary triage or repository orientation.
Build a compressed, token-efficient map of a repository with Repomix before analyzing or changing an unfamiliar codebase. Invoke explicitly with the host's Repomix skill command when repository-wide context is useful; do not use for a known single-file task.
Produce a compact, evidence-backed continuation brief for a fresh coding-agent task. Use before ending or replacing a long session; do not use as a substitute for completing a short active task.
Measure avoidable coding-agent input context from instruction files, installed skills, tool configuration, generated artifacts, and likely noisy outputs. Use when context limits, latency, or token cost are a concern; do not use for authentication tokens.
Save context for repositories, noisy output, long documents, or session history. Fast-pass short questions when optimization costs more than it saves.