Borrowing it
Nothing to install: this file belongs to sparq-org/sparq. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/sparq-org/sparq/main/.claude/skills/fused-decompress-parse/SKILL.mdgit clone --depth 1 https://github.com/sparq-org/sparqWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/sparq-org/sparq/fused-decompress-parse)<a href="https://agentmods.dev/skills/sparq-org/sparq/fused-decompress-parse"><img src="https://agentmods.dev/badge/skills/sparq-org/sparq/fused-decompress-parse/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/sparq-org/sparq/fused-decompress-parse"><img src="https://agentmods.dev/badge/skills/sparq-org/sparq/fused-decompress-parse.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00128 | $0.01659 |
| Opus 5 | $0.00064 | $0.00830 |
| Sonnet 5 | $0.00026 | $0.00332 |
| Haiku 4.5 | $0.00013 | $0.00166 |
Grade A, and why
fused-decompress-parse scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 10d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 109 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Fused decompress + parse for RDF ingest
[OPUS-4.8] Authored from measured research. Source of truth:
research/custom-parsers-baseline.md (the "Compressed ingest" section + the
post-fix measurements) and research/fast-ingestion.md. Verify before quoting.
The headline finding (and the bug it exposed)
A naïve streaming ingest (decoder → parser, one parse round per read())
was measured 3.5–5× SLOWER than just decompressing to RAM first — the
opposite of what streaming should cost. Two compounding causes, both fixed by
reusing code that already existed:
- Per-
read()flush bug.Graph::load_reader_parallelflushed a parse+sharded-merge round on everyread()call, and decompressors return small reads (gzip ~0.38 MB/read, zstd ~1.6 MB/read) into its 32 MiB buffer. The parallel parser+merge machinery is amortised for ~32 MiB blocks, not 0.4 MB ones. Fix: a producer thread that fills the full block acrossread()calls before handing it to the parser. - No pipelining. Decode and parse ran additively. Fix: run decode on its
own thread feeding a bounded channel, so it overlaps the rayon parse + dict
merge — the exact 3-stage pipeline that already existed in
build_external_ntriples_parallel(sparq-corelib.rs) and just wasn't used by the in-memory path.
After both fixes (same machine/harness), streaming ingest of the slice dropped by
~8–9× (gzip) and ~7× (zstd) and now matches or beats two-stage in every paired
run while never materialising the decompressed copy. (Absolute figures drift with
thermal state / machine — they live in research/custom-parsers-baseline.md and the
bench/parse/ harness; cite that, not baked numbers. The load-bearing fact is the
ratio + the "matches-or-beats two-stage" conclusion.)
The fusion bound — what "ideal" means
Ideal pipelined ingest = max(decode, parse+build), because a perfect pipeline
hides the smaller stage under the larger:
- gzip:
max(0.396, 0.296) ≈ 0.40 s - zstd:
max(0.140, 0.296) ≈ 0.30 s
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 10d ago First seen · 109 lines · 128 tokens per session scan A 6aefa792843b
fused-decompress-parse is a skill published in the GitHub repository sparq-org/sparq (12 stars, last pushed yesterday), licensed MIT. It adds 128 tokens to every session and 1,659 once invoked, about $0.0006 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
ax-rust-llm
Use when writing Rust code with axllm for using the generated Ax package, factory functions, package docs, examples, and API reference.
ax-rust-signature
Use when writing Rust code with axllm for string signatures, field descriptors, JSON schema output, validation, and typed tool argument shapes.
domain-ml
A guide to design constraints for machine-learning and AI applications written in Rust, including model inference and handling large numerical data.
rig
Guide for building LLM clients in Rust with rig (rig-core) — AgentBuilder, providers, tools, extractors, RAG, and streaming. Use when writing Rust LLM clients with rig or rig-core, building multi-provider or hybrid clients that mix Anthropic, OpenAI, Gemini, or local models, integrating ollama, LM Studio, or an…
ax-rust-gepa
Use when writing Rust code with axllm for GEPA, Pareto tradeoffs, reflection clients, metric budgets, optimizer state, and artifacts.
ort
ONNX Runtime in Rust via the ort crate (2.x): loading sessions, configuring CPU/CoreML/CUDA execution providers, tensor I/O with ndarray, async-safe spawnblocking wrapping, global thread-pool init, and debugging provider/opset issues.