← ClaudeAtlas

context-densitylisted

Agent context audits measure token cost, provenance, compression, typed-state drift, trigger overlap, and prompt/output contracts across AGENTS.md, prompts, skills, plugins, MCP/tool schemas, current-state artifacts, and agent handoffs.
Xopoko/plug-n-skills · ★ 11 · AI & Automation · score 77
Install: claude install-skill Xopoko/plug-n-skills
# Context Density Bundled commands use `$PLUGIN_ROOT` (`$env:PLUGIN_ROOT` in PowerShell; same path suffix) for the plugin root. Set it once: use the host's plugin-root variable when defined (Claude Code: `PLUGIN_ROOT="$CLAUDE_PLUGIN_ROOT"`), otherwise the absolute path of this plugin's root directory. ## Hard Rules - Preserve triggers, routing, safety, commands, output contracts, and validation proof before compressing wording. - Preserve trigger semantics, not just trigger phrases. - Descriptions should fire from task context, artifacts, source evidence, failures, and agent decision points; keep exact user wording only when it controls behavior, consent, or target binding. - Compress prose, not commitments: goals, constraints, decisions, IDs, paths, dates, warnings, evidence refs, safety boundaries, and behavior-critical exact wording need verbatim text or typed recovery pointers. - Treat raw logs, transcripts, reports, and source packs as evidence archives. Do not hot-load them; keep compact claims plus source refs. - Do not let retrieval, memory recall, or archived artifacts become authoritative state without provenance, confidence, and validation. - When companion artifacts can authorize actions or gate proof, seal their reviewed state in a typed state-commitment bundle and validate it; matching words do not prove semantic agreement. - Do not bury action-critical commitments in the middle of large hot/router files; keep anchors, source order, or explicit state pointer