/edikt:gov:score

Score the overall quality of compiled governance output. Measures context budget, LLM compliance across all directives, manual directive health, and surfaces the weakest links.

Usage

/edikt:gov:score
/edikt:gov:score --json

Arguments

Argument	Description
(none)	Human-readable governance quality report
--json	JSON output for CI integration

What it measures

Scoring reads the render manifest (.claude/rules/governance/manifest.yaml) and scores exactly the surfaces it lists — the ambient core, topic files, skill packages, and the directive index — rather than walking a directory, so it can't silently score a set nobody rendered. If the manifest is absent, gov:score says so and scores nothing rather than falling back to a walk.

Context budget

Total tokens across the ambient surfaces only — the ambient core (.claude/rules/governance.md) plus anything else loaded on every edit. Topic files load conditionally (only when a touched path matches their paths: frontmatter), and skill packages and the directive index are trigger- or write-time-loaded, not ambient cost — none of those count against this budget, since counting them would report a cost nobody actually pays on a typical edit.

Budget	Rating
<1000 tokens	Lean
1000-2000	OK
2000-4000	Heavy
>4000	Warning

LLM compliance metrics

Each directive is scored on:

• Token specificity — literal code tokens (backtick-wrapped identifiers, function names). High = 3+, Medium = 1-2, Low = 0.
• MUST/NEVER — hard constraint language present.
• Grep-ability — can compliance be checked with a shell command.
• "No exceptions." — present on invariant directives with absolute language.

Manual directive health

Manual directives bypass compile quality checks. This command scores them to the same standard — flags soft language, missing references, and conflicts with auto-generated directives.

Reminders and checklist

Counts items in each topic file's ## Reminders and ## Verification Checklist sections — these sections live per topic file, not in the ambient core, which carries neither. Flags when empty (missing) or when exceeding caps (10 reminders, 15 checklist items).

Example output

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 GOVERNANCE QUALITY REPORT
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Context budget: 847 tokens — OK
Sources: 7 ADR, 2 INV, 3 guidelines
Directives: 18 auto + 3 manual = 21 total

LLM Compliance:
  Token specificity:  18/21 high
  MUST/NEVER:         19/21 (90%)
  Grep-able:          15/21 (71%)
  "No exceptions.":   4/5 invariant directives
  Average score:      7.8/10

Reminders: 6 items
Checklist: 8 items

Manual Directive Health:
  Passing quality:    1/3
  Needs rewrite:      2/3

Overall: 7.8/10
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

When to run

• After /edikt:gov:compile to verify output quality
• In CI to enforce a minimum governance score
• Periodically to catch governance drift as the project grows

Related commands

• /edikt:gov:compile — generate the governance files this command scores
• /edikt:gov:review — review governance for contradictions and language quality
• /edikt:invariant:review — per-invariant quality scoring (human-side + directive-side)
• /edikt:adr:review — per-ADR quality scoring