Conciseness

The body is lean and factual throughout — terse bullet lists ('Real network support: yes.'), one short code snippet, and no explanation of concepts Claude already knows. Every section carries only information specific to the axllm package, and no version numbers or dates appear in the body.

Actionability

The Core Pattern snippet is concrete but not executable as written since `llm` is never constructed or defined, and the 'Relevant API Surface' is a bare list of names with no call shapes. Most importantly, the skill's stated core task — 'Choose between contextMap, contextPolicy, optimization, and recall' — is given no decision criteria or concrete selection guidance, only guardrails pointing to example files that are not in the bundle.

Workflow Clarity

Guidance exists ('Start from package examples for exact native syntax', 'Use `no-key` examples for deterministic local checks') but is presented as unsequenced bullets with no ordered procedure or decision flow for the feature-routing task. This is not a destructive or batch skill so the validation cap does not apply, but there is no explicit sequence to reach anchor 4.

Progressive Disclosure

The under-50-line body is well organized into clear sections (When To Use, Package Facts, Core Pattern, API Surface, Guardrails) with nothing inlined that belongs in a separate file. However, the referenced materials — 'API.md', 'axir-api.json', 'axir-capabilities.json', 'examples/' — do not exist in the skill bundle and are named rather than linked, so navigation is asserted but not resolvable.

Specificity

It names the domain ('writing C++ code with `axllm`') and concrete feature areas ('context maps, trajectory context policy, offline optimization (ACE/GEPA), and memory recall'), but the only action framing is the abstract 'deciding between' — there is no list of concrete actions performed. It sits above anchor 2 (features are specific, not generic) but below anchor 4 (no several specific actions are enumerated).

Completeness

The 'when' is explicit ('Use when writing C++ code with `axllm`') and the 'what' is present ('deciding between context maps, trajectory context policy, offline optimization (ACE/GEPA), and memory recall'), but the what is embedded inside the when-clause rather than stated as a standalone capability statement with its own concrete trigger phrases. This matches anchor 4 and falls short of anchor 5's fully explicit both-parts form.

Trigger Term Quality

Terms like 'axllm', 'ACE/GEPA', 'context maps', and 'long-context agents' are what a user of this niche package would plausibly say, so this is above anchor 2. But the phrasing is jargon-dense and misses natural variations a user might actually type (e.g., 'prompt optimization', 'context selection', 'C++ agent'), keeping it below anchor 4's 'good keyword coverage'.

Distinctiveness Conflict Risk

The description occupies a clear niche — C++ code with the specific generated package `axllm` for Ax-specific concerns (AxIR, ACE/GEPA, trajectory context policy). Triggers are distinct and it is very unlikely to fire for any unrelated skill, matching anchor 5.