Conciseness

The body is lean and assumes competence: terse fact lists ('Real network support: yes. Scripted no-key transport support: yes.'), a compact 'Relevant API Surface' list, and no explanations of concepts Claude already knows. Every section carries information, matching the 'every token earns its place' anchor.

Actionability

The Core Pattern ('auto helper = axllm::agent("question:string -> answer:string");') is concrete C++ but depends on an undefined 'llm' variable, and 'When To Use' says 'Choose between contextMap, contextPolicy, optimization, and recall for a task' without any criteria for how to choose — the skill's core promised function has no executable decision guidance. This matches the 'some concrete guidance but incomplete; missing key details' anchor rather than 4, where guidance would be mostly executable.

Workflow Clarity

The routing workflow is a rough bullet list ('Choose between...', 'Avoid mixing persistent corpus orientation with within-run compaction', 'Route long-context agent work...') with no decision sequence, no checkpoints, and no validation steps; the Guardrails section partially compensates with do/don't rules. This sits between anchor 2 (many gaps, validation absent) and anchor 3 (sequence present but checkpoints missing) — closer to 3 because the guardrails act as implicit checkpoints, but not 4 since the decision flow for the skill's primary task is undefined.

Progressive Disclosure

The skill is under 50 lines with clear, well-organized sections and no external bundle files; references to 'API.md', 'axir-api.json', 'axir-capabilities.json', and 'examples/' are package artifacts clearly signaled as such and one level deep. Per the simple-skill scoring note, this matches the top anchor.

Specificity

It names the domain ('writing C++ code with `axllm`') and enumerates four concrete decision targets ('context maps, trajectory context policy, offline optimization (ACE/GEPA), and memory recall'), but the only action verb belongs to the user's activity ('writing C++ code'), not the skill — what the skill actually does is implied. This matches anchor 3 (domain and concrete items, but not comprehensive) rather than 4, whose examples list several specific actions the skill itself performs.

Completeness

The 'when' is explicit ('Use when writing C++ code with `axllm` for...'), and a 'what' is present ('deciding between context maps, trajectory context policy, offline optimization (ACE/GEPA), and memory recall') but carried by an indirect prepositional clause rather than an explicit skill-action statement. This sits between anchor 3 (what clear, when weak) and anchor 4 (both present, one could be more specific) — above the midpoint because the when is fully explicit, but short of 5 because the what lacks a direct action verb.

Trigger Term Quality

Good natural keyword coverage: 'C++', 'axllm', 'context maps', 'trajectory context policy', 'offline optimization', 'ACE', 'GEPA', 'memory recall', 'long-context agents', with synonym variants (ACE/GEPA). Missing a few broader phrasings users might naturally use (e.g., 'context window', 'prompt optimization'), so it falls just short of the comprehensive synonym coverage of anchor 5.

Distinctiveness Conflict Risk

The trigger terms are highly specific to a niche ('C++', 'axllm', 'ACE/GEPA', 'trajectory context policy'), creating a clear niche with distinct triggers and minimal risk of firing for unrelated skills — matching the top anchor exactly.