Conciseness

The body is telegraphic and assumes competence — no padding, no explanation of concepts Claude already knows ('Never run both: every span doubles', 'Silence alone proves nothing'). Docked from 5 for time-sensitive information scattered outside any 'old patterns'/'deprecated' section: 'tested 0.14.25', native otel '0.7.0', 'pin llama-index-core<0.14.25' pending OpenInference PR #3841. Not 3 because there is no unnecessary explanation anywhere.

Actionability

Fully executable and copy-paste ready: a complete verbatim `tracing.py` processor, exact pip pins, an exact `OTEL_EXPORTER_*` env block, and concrete snippets for `using_session(...)`, `instrument_tags(...)`, and `stream_options: {include_usage: true}`. Matches the 5 anchor; not 4 because the common cases (install, init, session id, flush, verify) each come with ready code or commands.

Workflow Clarity

Clear 0-8 step sequence (Detect → Key/region → Install+init → Session id → Content → Tools/sub-agents → Tokens → Flush → Verify) with an explicit verification checklist in Step 8 and feedback loops (401 → try the other region; 'Silence alone proves nothing' → console-span fallback check; 'Ask before removing'). Matches the 5 anchor's explicit validation steps and checklists; not 4 because every checkpoint is present, including a no-Maple-access fallback.

Progressive Disclosure

Well-organized single-file structure with clearly numbered sections and no nested references — but the ~40-line Edge cases tail (native otel 0.7.0 details, CondensePlusContextChatEngine span layout, OpenRouter Broadcast nesting, streaming query engine pin) is situational material inlined in SKILL.md that fits a one-level reference file. Matches the 4 anchor ('good structure; most content appropriately placed; minor organization gaps'); not 5 because nothing is offloaded, and not 3 because the core procedure legitimately belongs inline and navigation is easy.

Specificity

Lists multiple specific concrete capabilities — 'OpenInference LlamaIndex instrumentor with GenAI output, a conversation id per chat, agent names and one span per model call, so each conversation is one Maple Agent Session with transcript, tools and tokens' — comprehensively covering what the skill delivers. Matches the 5 anchor ('multiple specific concrete actions; comprehensive coverage'); not 4 because no meaningful capability is left out.

Completeness

Explicitly answers 'what' (instrument LlamaIndex agents so each conversation is one Maple Agent Session with transcript, tools and tokens) and 'when' via an explicit 'Triggers on ...' clause with concrete phrases. Matches the 5 anchor exactly; the 'Use when...' cap at 3 does not apply.

Trigger Term Quality

Five natural trigger phrases users would actually say ('trace my llamaindex agent', 'add Maple to llamaindex', 'agent sessions for llamaindex', 'OpenTelemetry for llamaindex', 'llama-index observability'), including the llamaindex/llama-index spelling synonyms. Matches the 5 anchor's comprehensive coverage with synonyms; not 4 because trace/add/sessions/observability/OTel phrasings are all covered.

Distinctiveness Conflict Risk

Clear niche locked to both 'Maple' and 'LlamaIndex' with triggers that mix the two ('add Maple to llamaindex'); a user saying these phrases unambiguously wants this skill. Minimal conflict risk; not 4 because even the more generic phrases ('llama-index observability') are within this skill's legitimate scope of handling existing OTel setups.