Conciseness

The execution phases are efficient, but the 'What is XXE' tutorial restates knowledge Claude already has, and the 8 'Patterns That Prevent XXE' snippets are largely duplicated by the SECURE halves of 'Vulnerable vs. Secure Examples' (with safe patterns appearing a third time in Phase 1's 'What to skip'). Mostly efficient with some unnecessary explanation and tightening opportunities — anchor 3; not 2 because the majority of the body (phase instructions, flag/skip lists) earns its tokens.

Actionability

Guidance is fully executable: grep-able per-language API lists ('etree.fromstring(...)', 'DocumentBuilderFactory.newInstance()'), copy-paste subagent prompt templates with exact output formats, and a ready curl XXE payload with Content-Type header. Matches anchor 5; not 4 because the common cases are covered end-to-end with no gaps.

Workflow Clarity

The three phases are explicitly ordered ('Phase 2 must run AFTER Phase 1 — it depends on the recon output'), with an early-exit validation checkpoint on zero recon findings, a worked batching procedure ('8 sites → 3 batches'), a merge procedure, and intermediate-file cleanup. This batch-operation skill has explicit checkpoints and error-recovery classification ('Needs Manual Review'), matching anchor 5; not 4 because validation is present at every phase boundary.

Progressive Disclosure

Section headers organize the body well, but ~290 lines of per-language reference material (prevention patterns, vulnerable/secure examples) are inlined in a single 570-line file with no bundle files, and Phase 2's manual '[TECH-STACK EXAMPLES]' selection is exactly what per-stack reference files would automate. Some structure with content that should be separate kept inline matches anchor 3; not 4 because there is no reference-file split at all, not 2 because sections are clearly headed and navigable rather than a wall of text.

Specificity

The description lists multiple concrete actions — 'recon (find XML parsing sites without external-entity hardening)', 'batched verify (trace user input to each site in parallel subagents, 3 sites each)', 'merge (consolidate batch results)' — plus explicit input ('Requires sast/architecture.md') and output ('Outputs findings to sast/xxe-results.md') artifacts. This matches the comprehensive-coverage anchor; it is not 4 because there are no gaps in what/mechanics/inputs/outputs.

Completeness

It explicitly answers both questions: 'what' is concrete (three named phases with artifacts) and 'when' is an explicit trigger clause — 'Use when asked to find XXE or XML injection bugs'. This matches the anchor-5 example structure; it is not 4 because the 'when' is explicit with concrete trigger phrases rather than vague.

Trigger Term Quality

It includes the natural phrases users would say — 'XXE', 'XML External Entity', 'find XXE or XML injection bugs' — but omits common synonyms such as 'external entity injection', 'XML parser vulnerability', or file-extension hooks. Good coverage with a few natural terms missing fits anchor 4; not 5 because synonym coverage is incomplete, not 3 because the primary natural terms are all present.

Distinctiveness Conflict Risk

'XXE' is a clear niche with distinct triggers, and 'Requires sast/architecture.md (run sast-analysis first)' explicitly disambiguates it from the sibling analysis skill. Minimal conflict risk matches anchor 5; it is not 4 because no closely related skill would trigger on these terms in preference to this one.