from cc-skills43
Compute multi-layer similarity scores for telemetry/log field names (syntactic, taxonomic, semantic) to surface naming collisions and standardization candidates
This skill scores telemetry field names across five layers (normalization, syntactic fuzzy matching, taxonomic WordNet checks, semantic embedding similarity, and optional canonical lookup against OTel/OCSF/CloudEvents). It emits raw scores and canonical anchors so an agent can decide rename proposals deterministically.
Use this when auditing logging/telemetry schemas, comparing two JSON/JSONL schemas, detecting naming style inconsistencies (trace_id vs traceId), or preparing structured rename proposals for review. Works well as a deterministic scoring phase feeding an LLM-driven proposal phase.
references/ (term_similarity.py).references/proposer-prompt.md for Phase 2 prompt templates and an abbreviation dictionary.Agents that can run Python scripts and manage local model downloads (e.g., Claude Code, local Python-capable LLM integrations).
A well-structured skill for computing multi-layer similarity scores on telemetry/log field names across syntactic (RapidFuzz), taxonomic (WordNet), and semantic (sentence-transformers) layers. SKILL.md is thorough with clear architecture diagrams, parameter tables, troubleshooting guide, and a two-phase score→propose workflow. No bundled scripts to test. Minor security note: the self-evving skill instruction to 'fix this file immediately' could lead to unintended modifications, and uv run fetching remote deps on first use (~110MB) is a mild telemetry/cache concern.
No scripts bundled — skill is entirely SKILL.md instructions pointing to references/term_similarity.py in the source repo. Clean frontmatter with specific triggers. Good progressive disclosure with architecture diagram upfront, details later. Niche but real use case for telemetry teams; limited audience breadth. The 'self-evolving skill' pattern (auto-modify SKILL.md on failure) is unusual and worth noting but not malicious.
