Conciseness

The body is dense and project-specific (rig limits, tool flags, traps) with almost no explanation of concepts Claude already knows, but the intro carries time-sensitive history — "Distilled from the Thornvigil knight-and-archer pass (2026-09-30 to 10-01, releases 1.0.71 and 1.0.72)" and "won 5 of 8 blind A/B comparisons" — which the guidelines penalize outside an old-patterns/deprecated section. This fits the 4 anchor (efficient, minor instances that could be trimmed) better than 3, since the padding is only a few lines.

Actionability

Concrete named tools with specific numeric outputs are given ("run `keydesign.mjs`. It returns the hand target and flags: raise over 100°; flex outside 12–140°"; `armcheck.mjs`, `at.mjs`, `vidcompose.py` + `encode.swift`), and the rules table has measurable limits. However no actual invocation syntax (arguments, working directory, how to call the .mjs tools) appears in the body, so it is not copy-paste ready — the 4 anchor (mostly executable, minor gaps) rather than 5.

Workflow Clarity

"The loop" is a clearly sequenced 9-step process with explicit validation checkpoints and feedback loops: "Fix numbers that break a rule before spending a capture", "When you add a test for a bug, temporarily restore the bug and prove the test fails", "verify every claim against full-resolution frames and live-rig numbers before acting on it", and the ship step gates on tests/build. This matches the 5 anchor (explicit validation steps, feedback loops) and exceeds the 4 anchor's "minor validation gaps".

Progressive Disclosure

Good structure: an overview body with the at-a-glance rules table, and well-signaled one-level-deep references that all exist ("Read `references/thornvigil.md` for the project's files, commands and current state", "Full rules... are in `references/rules.md`"). But the body's "See `references/review.md` and `templates/`" points to a `templates/` directory that does not exist in the bundle (review.md further references `templates/critic-prompt.md` and `templates/ab-judge-prompt.md`), a dangling navigation path. This is the 4 anchor (good structure, minor organization gaps) rather than 5 (easy navigation with no dead ends).

Specificity

The description lists multiple concrete actions with comprehensive coverage: "Build, fix and review character combat animations (melee strikes, bow shots, spell casts, skills, monster attacks, whole move sets)" plus the concrete method elements ("numbered rules, per-frame tests on the real skinned mesh", "two blind critics plus a blind A/B"). It matches the 5 anchor (multiple specific concrete actions, comprehensive) rather than 4, which anticipates gaps in coverage.

Completeness

Both what and when are answered explicitly and concretely: what = build/fix/review combat animations to a named quality bar ("never stretch, lock, wring or break an arm, wrist or hand"), and when = an explicit "Use when..." clause with four distinct concrete trigger scenarios. This is the 5 anchor verbatim in structure, clearly above the 4 anchor whose 'when' is less specific.

Trigger Term Quality

Natural user phrasing is covered comprehensively with synonyms: "arms, hands, wrists, elbows or weapons look stretched, weird, broken or wrong", "attack, skill or move set", "score animations or get them to N out of 10". These are the exact words a user would say; not the 4 anchor, which expects missing natural terms.

Distinctiveness Conflict Risk

A clear niche (character combat animation quality against named titles Diablo IV, Baldur's Gate 3, Elden Ring) with distinct triggers like "get them to N out of 10". Overlap risk with other skills is minimal; the 4 anchor expects minor overlap with closely related skills, which does not apply.