Arxiv Summarizer Orchestrator
PassAudited by ClawScan on May 1, 2026.
Overview
This instruction-only skill transparently coordinates arXiv collection, downloading, summarization, and reporting, with no artifact evidence of hidden credential use, exfiltration, or destructive behavior.
Before installing, confirm you also trust the three required arXiv sub-skills, run the workflow in a dedicated output directory, keep parallelism within the documented limits, and only enable scheduled execution if you want recurring automated paper collection.
Findings (5)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
A malicious or unusual paper could contain text that looks like instructions; the agent should ignore such text except for summarization.
The skill intentionally has the model read downloaded paper content. Academic source/PDF text is external content and should be treated as data, not as instructions to the agent.
Model reads content and manually writes `<paper_dir>/summary.md` by reference format, in `language`.
Use this skill only for summarization, and instruct the agent to treat paper contents, source files, and PDFs as untrusted reference material.
Using the skill will cause local commands from the related sub-skills to run, fetch arXiv data, and create or update files in the chosen run directory.
The workflow directs the agent to run local helper scripts for fetching, downloading, and reporting. This is central to the skill's purpose and is clearly disclosed.
Run `arxiv-search-collector/scripts/fetch_queries_batch.py` with the plan file ... Run one batch artifact download ... `download_papers_batch.py`
Run it in a scoped output directory and review the related sub-skills' scripts before allowing broad or scheduled runs.
The safety of the full workflow also depends on the provenance and behavior of those referenced sub-skills.
This top-level skill relies on separately installed sub-skills whose code is not included in this artifact set.
Before using it, install and enable these three sub-skills: `arxiv-search-collector`, `arxiv-paper-processor`, `arxiv-batch-reporter`.
Install only trusted, reviewed versions of the three required sub-skills and verify their permissions and scripts separately.
If parallel workers are mis-scoped, they could overwrite or mix outputs between paper directories.
The skill uses multiple subagent worker runs over shared workflow output, while also documenting a per-paper ownership boundary.
In parallel mode, run multiple `arxiv-paper-processor` instances in batches ... Subagent workers should only own one paper directory each to avoid file conflicts.
Keep the default concurrency limits and ensure each worker is assigned exactly one paper directory.
If scheduled, it may repeatedly fetch papers and generate files without a fresh manual prompt each time.
The artifact explicitly supports recurring automated execution, but does not install persistence or hide background behavior.
This orchestrator is suitable for cron/scheduled execution in OpenClaw: Frequency examples: daily, weekly, monthly.
Only configure scheduling deliberately, choose a clear output location, and monitor generated runs.