Content
77%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is an exceptionally well-sequenced, validation-heavy workflow: every phase has concrete commands, numeric gate targets, hard ordering rules, bounded review loops, and resume state — a model of gated execution. Its weaknesses are duplicated rule statements across sections and a progressive-disclosure gap: many files the workflow depends on (templates/, reference poster sets, NOTICE/LICENSES, shared-references) are not present in the shipped bundle, leaving several steps non-executable from this install alone.
Suggestions
Ship (or verify) the referenced bundle files: templates/ (README.md, COMPONENTS.md, tokens/generic.json), references/good/ and references/bad/ for calibration, NOTICE.md, and LICENSES/posterly-MIT.txt — Phase 0.5, 3, and 5 steps currently point at files absent from the bundle.
Move the detailed gate specifications (figure-area bands, the visual-score formula/caps, fix vocabulary detail) into a one-level-deep reference file and keep SKILL.md as overview plus commands, reducing duplication with the Constants and Key rules sections.
Give the codex review calls in Phases 1 and 6 a concrete invocation template (tool name plus a minimal prompt skeleton) so the cross-model steps are executable without relying on the missing ../shared-references/reviewer-backends.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and operational with no explanations of concepts Claude already knows, but there is trimmable material: the 12-line predecessor-failure narrative in "Why this skill exists" and rules restated 2–3 times across Constants, Phases, and "Key rules" (e.g., "Never invent paper numbers or figures", gates-before-aesthetics, two-hue discipline). This matches 'efficient; minor instances of over-explanation' rather than the lean 5 anchor. | 4 / 5 |
Actionability | Mostly executable: copy-paste commands with exact arguments ("python3 \"$SKILL_SCRIPTS/run_gates.py\" poster_html/poster.html --tokens <pack.json> --manifest ... --report ...", "pdftoppm -r 100 ... -png -f 1 -l 1", "poster_check.py verify-final ... --max-size-mb 20") and precise numeric targets (spread < 5 px, canvas-fill ∈ [95, 101] %, 550–850 words). Not a 5: the codex review calls (Phase 1/6) never show concrete invocation syntax, and component swaps rely on templates/COMPONENTS.md, which is not present in the shipped bundle. | 4 / 5 |
Workflow Clarity | Phases 0–7 are clearly sequenced with 🚦 checkpoints, hard validation gates before any review, explicit error-recovery feedback loops ("fix → re-run Phase 4 gates → re-score", "≤ 3 issues per round, ≤ 3 rounds", escalation to template/canvas re-choice), resume state via POSTER_STATE.json, and a final verify-final step — squarely the anchor for clear sequence with explicit validation and feedback loops. | 5 / 5 |
Progressive Disclosure | Section headers are clear and references are one level deep and well-signaled, but scored against the actual bundle: cited paths — templates/README.md, templates/COMPONENTS.md, templates/tokens/generic.json, references/good/ and references/bad/, NOTICE.md, LICENSES/posterly-MIT.txt, and ../shared-references/{reviewer-backends,taste-calibration,review-tracing}.md — are absent from the bundle (only scripts/ ships), so navigation breaks; and detailed gate/scoring tables (figure-area bands, score formula and caps) are inlined where a reference file would be appropriate. More than the 'minor organization gaps' of the 4 anchor. | 3 / 5 |
Total | 16 / 20 Passed |