context-dietlisted

Use when the user asks to audit coding-agent context, reduce token or context waste, diagnose Context Diet, start or stop the local service, explain a policy decision, run a benchmark, retrieve exact context, understand why a block stayed native, or troubleshoot Codex or Claude Code setup.
mSolstice/ContextDiet · ★ 0 · AI & Automation · score 55

Install: claude install-skill mSolstice/ContextDiet

# Context Diet Operate the local Context Diet engine for Codex and Claude Code. ## Rules - This skill operates a local engine and gateway for future provider requests. - It cannot compress or transform the request already invoking this skill. - Safe mode is the default. - Exact, current, action-critical and secret-sensitive material stays protected. - Benchmark results may be positive, neutral or negative. - Never request, print or store provider credentials. ## Workflow 1. Classify the task: audit, reduce, diagnose, start, stop, explain, benchmark, retrieve or setup. 2. Read only the relevant reference below. 3. Prefer local CLI commands and deterministic repository scripts. 4. Preserve safety boundaries; do not weaken policy rules for savings. 5. Report measured versus estimated results and any blocked action. ## References - `references/OPERATIONS.md`: start, stop, status, doctor, service routing. - `references/SAFETY.md`: policy decisions, protected material, exact retrieval boundaries. - `references/BENCHMARKING.md`: audits, benchmarks, projections, share cards. - `references/TROUBLESHOOTING.md`: failures, degraded mode, setup symptoms. - `references/CODEX_SETUP.md`: Codex-only setup details. - `references/CLAUDE_CODE_SETUP.md`: Claude Code-only setup details.