Conciseness

The body is a ~128-line, 23KB wall of dense prose that exhaustively inlines every option's edge cases (e.g., the 30+ line 'Provider Forward Options' section and the 'Astra Session Work' section), with the phrase 'as in TypeScript' repeated roughly twenty times as padding. It is noticeably verbose with many sections that should be trimmed or moved out, though it does not explain concepts Claude already knows, keeping it above the 1 anchor.

Actionability

There is real concrete guidance — the core-pattern code block, exact option spellings, defaults ('maxSteps default 25', 'maxRetries default 3'), method signatures like `ax(..., sample_count=N, result_picker=callback)` and `add_field_processor(field, fn)`, and exact error strings — but it is incomplete: `llm` in the core pattern is never constructed or explained, and no complete executable example shows tools, streaming, or multi-sampling end to end. This matches the 3 anchor: some concrete guidance, key details missing.

Workflow Clarity

'When To Use' bullets and the 'Guardrails' section give a rough orientation (start from package examples, use no-key examples for deterministic local checks), but there is no sequenced multi-step process with explicit validation checkpoints — the validation hint is a single implicit guardrail line. This fits the 3 anchor: sequence present but checkpoints missing or implicit.

Progressive Disclosure

The skill bundle contains no `references/`, `scripts/`, or `assets/` directories, so the paths the body relies on (`API.md`, `axir-api.json`, `axir-capabilities.json`, `examples/`, `src/examples/python/generation/`) do not resolve, and the bulk of the body is an inlined API/behavior specification that clearly belongs in separate reference files. This matches the 2 anchor: content that belongs in separate files is inlined and references are effectively buried/dangling, despite the section headers keeping it above a 1.

Specificity

The description enumerates many concrete capability areas — 'AxGen programs, forward calls, indexed multi-sampling, result pickers, streaming, tools, assertions, traces, usage, and output parsing' — giving broad, specific coverage of the package. It falls short of the 5 anchor because the items are feature-area nouns under a single generic action ('writing Python code with `axllm`') rather than multiple distinct concrete actions like the anchor's 'extract text, fill forms, merge documents, convert pages'.

Completeness

Both what ('writing Python code with `axllm` for AxGen programs, forward calls, ... output parsing') and when ('Use when writing Python code with `axllm` ...') are explicit. It sits below 5 because the when-clause simply restates the what rather than naming distinct trigger situations or user mentions (compare the 5 anchor's 'or when the user mentions PDFs, forms, or document extraction').

Trigger Term Quality

Terms like 'Python code', 'axllm', 'AxGen', 'forward calls', 'streaming', 'multi-sampling', 'result pickers', and 'output parsing' are exactly what a user of this library would say, giving good keyword coverage. A few natural variations are missing (e.g., 'structured generation', 'structured output', 'MCP', 'caching'), so it does not reach the comprehensive-with-synonyms 5 anchor.

Distinctiveness Conflict Risk

The niche markers '`axllm`' and 'AxGen' are highly distinctive package names unlikely to appear in any other skill's context, and the enumerated feature areas further narrow the trigger. This matches the 5 anchor: a clear niche with distinct triggers and minimal conflict risk.