| name | evaluate-context-engineering |
| description | Context-engineering audit (C1-C6) of skills and always-loaded rules. Use when trimming context bloat, auditing a plugin, or after a new Claude model ships. |
| args | [plugin | plugin/skill | --always-loaded] [--judge] |
| allowed-tools | Bash(python3 *), Bash(bash *), Read, Glob, Grep, Task, TodoWrite |
| argument-hint | git-plugin/git-commit --judge |
| model | opus |
| created | "2026-07-26T00:00:00.000Z" |
| modified | "2026-07-26T00:00:00.000Z" |
| reviewed | "2026-07-26T00:00:00.000Z" |
/evaluate:context-engineering
Audit a plugin, a single skill, or the always-loaded surface against the six
context-engineering shifts Anthropic named for Claude 5 generation models, and
report a C1–C6 card.
Two channels, deliberately kept apart:
| Channel | What it is | Cost |
|---|
| M — mechanical | check-context-engineering.py, deterministic proxies over the tree | free |
| J — judgment | an isolated agent scoring the frozen 1–5 rubric with quoted evidence | one subagent, --judge only |
Channel M ranks candidates; it cannot tell a load-bearing safety constraint from
a restatement of a model default. Channel J makes that call. Never blend the
two into one score — a mechanical proxy and a judgment answer different
questions, and averaging them hides which one is driving a recommendation.
When to Use This Skill
| Use this skill when... | Use alternative when... |
|---|
| Auditing whether a skill or rule is written for the old context rules | Checking structure/frontmatter/size lint → scripts/plugin-compliance-check.sh |
| A new Claude model shipped and the corpus needs re-reading | Measuring whether a skill still works → /evaluate:matrix |
Deciding what to cut from CLAUDE.md and .claude/rules/ | Actually migrating rules into skills → agent-patterns-plugin:meta-context-diet |
| You want ranked, evidence-backed cut candidates | Checking a skill reads clearly to a fresh agent → /evaluate:legibility |
The six dimensions
| Dim | Shift | What the scanner measures |
|---|
| C1 | rules → judgment | Constraint markers per 1k chars, and how many sit next to a safety/irreversibility word |
| C2 | examples → interface design | Fenced-block share of the body; whether args enumerates its alternatives |
