Conciseness

The body is dense and imperative with essentially no explanation of concepts Claude already knows — every section is commands, defaults, or constraints. It is penalized one notch because time-sensitive version information ("GPT‑5.2 Pro", "gpt-5.2-pro") appears throughout rather than in a deprecated/old-patterns section, and a few sections (remote browser host, exhaustive prompt pattern) run slightly long; anchor 4 rather than 5.

Actionability

Guidance is fully executable and copy-paste ready: complete commands like `npx -y @steipete/oracle --engine browser --model gpt-5.2-pro -p "<task>" --file "src/**"`, `oracle status --hours 72`, and exact mkdir/redirect recipes for the manual fallback. Defaults, excludes, and hidden flags are all specified concretely, matching anchor 5.

Workflow Clarity

The "Golden path" gives a clear numbered sequence with an explicit pre-send validation step ("Preview what you're about to send (`--dry-run` + `--files-report`)"), token-budget checkpoints, and an error-recovery loop for the main failure mode ("If the run detaches/timeouts: reattach to the stored session (don't re-run)" with the reattach commands). This matches anchor 5's sequence + checkpoints + feedback loop.

Progressive Disclosure

No bundle files exist (references/, scripts/, assets/ are absent) and the single SKILL.md is well-organized with clear section headers covering one coherent topic, so there are no buried or nested references. It falls just short of anchor 5 because at ~115 lines, secondary material (remote browser host setup, the exhaustive-prompt pattern, session management details) could arguably live in one-level-deep reference files — minor organization gaps place it at 4.

Specificity

Quotes several concrete actions — "bundle a prompt plus the right files", "get a second-model review (API or browser)" — with the named @steipete/oracle CLI domain. It lists multiple specific actions but coverage is not comprehensive (attachment selection, sessions, cost behavior are absent), matching anchor 4 rather than 5.

Completeness

The "what" is clear (bundles prompt + files into a second-model review via API or browser) and an equivalent "when" clause exists ("for debugging, refactors, design checks, or cross-validation"), so it avoids the missing-trigger cap of 3. It lacks the explicit "Use when..." phrasing and could be more specific, matching anchor 4 rather than 5.

Trigger Term Quality

Natural trigger terms include "debugging, refactors, design checks, or cross-validation" plus "second-model review" — phrases a user would plausibly say. Common synonyms like "second opinion", "code review", or "another model" are missing, so it fits anchor 4 (good coverage, a few natural terms missing) rather than 5.

Distinctiveness Conflict Risk

The mechanism (bundling a prompt plus files for a second model via the @steipete/oracle CLI) carves out a distinct niche with minimal confusion risk. Terms like "debugging" and "design checks" mildly overlap with general code-review skills, placing it at anchor 4 rather than 5.