ai-news-pipeline Security Report — Trusted | ClawSafe

5 /100

ai-news-pipeline

Self-contained Chinese and international AI news RSS capture and AI-enriched report generation workflow

A clean AI news RSS pipeline that collects feeds, generates AI-enriched reports, and writes outputs — all functionality matches documentation with no malicious indicators.

Skill Nameai-news-pipeline

Duration56.4s

Enginepi

✓

Safe to install

Approve for deployment. No security concerns identified.

Findings 2 items

Severity	Finding	Location
Low	base64 HTTP Basic Auth usage not declared in SKILL.md Doc Mismatch collect_feeds.py uses base64.b64encode to build HTTP Basic Auth headers for feeds that have username/password credentials (line 160-161). SKILL.md does not mention this capability. However, the scope is narrow and benign — it is standard HTTP auth for RSS feeds, not credential exfiltration. `basic_token = base64.b64encode(f"{source.username}:{source.password}".encode("utf-8")).decode("ascii")` → Document in SKILL.md that the skill supports authenticated RSS feeds via HTTP Basic Auth.	`scripts/collect_feeds.py:160`
Low	subprocess pipeline chaining not declared in SKILL.md Doc Mismatch All three entrypoints (run_capture_only.py, run_report_only.py, run_full_workflow.py) use subprocess.run to invoke other scripts within the skill. SKILL.md describes each entrypoint independently but does not mention that they internally use subprocess. This is legitimate pipeline orchestration but is a minor documentation gap. `subprocess.run(command, check=True, cwd=workspace, env=env)` → Add a note in SKILL.md that entrypoints may chain into other bundled scripts via subprocess.	`scripts/run_capture_only.py:27`

Resource	Declared	Inferred	Status	Evidence
Filesystem	`WRITE`	`WRITE`	✓ Aligned	Scripts write to configurable workspace subdirectories (data/, reports/, state/,…
Network	`READ`	`READ`	✓ Aligned	RSS feed collection via urllib (collect_feeds.py:fetch_feed) and ARK API calls (…
Shell	`NONE`	`WRITE`	✓ Aligned	Entrypoints use subprocess.run to chain scripts (run_capture_only.py:27, run_rep…
Environment	`NONE`	`READ`	✓ Aligned	os.environ.copy() passed to subprocess; ARK_API_KEY, ARK_API_BASE, ARK_MODEL, AR…
Skill Invoke	`NONE`	`NONE`	—	No cross-skill invocation detected.
Clipboard	`NONE`	`NONE`	—	No clipboard access.
Browser	`NONE`	`NONE`	—	No browser automation.
Database	`NONE`	`NONE`	—	No database access.

3 findings

🔗

Medium External URL 外部 URL

http://www.w3.org/2005/Atom

scripts/collect_feeds.py:87

🔗

Medium External URL 外部 URL

http://purl.org/rss/1.0/modules/content/

scripts/collect_feeds.py:89

🔗

Medium External URL 外部 URL

https://ark.cn-beijing.volces.com/api/v3

scripts/generate_company_report.py:36

File Tree

12 files · 86.5 KB · 2575 lines

Python 8f · 2255L Markdown 2f · 314L YAML 1f · 4L Text 1f · 2L

├─ ▾ 📁 agents

│ └─ 📋 openai.yaml YAML 4L · 490 B

├─ ▾ 📁 references

│ └─ 📝 commands.md Markdown 179L · 5.4 KB

├─ ▾ 📁 scripts

│ ├─ 🐍 brief_writer.py Python 314L · 11.0 KB

│ ├─ 🐍 collect_feeds.py Python 532L · 17.2 KB

│ ├─ 🐍 generate_company_report.py Python 560L · 18.7 KB

│ ├─ 🐍 generate_international_report.py Python 543L · 20.0 KB

│ ├─ 📄 requirements.txt Text 2L · 22 B

│ ├─ 🐍 run_capture_only.py Python 36L · 921 B

│ ├─ 🐍 run_full_workflow.py Python 65L · 1.8 KB

│ ├─ 🐍 run_report_only.py Python 69L · 2.1 KB

│ └─ 🐍 time_window.py Python 136L · 4.2 KB

└─ 📝 SKILL.md Markdown 135L · 4.5 KB

Dependencies 2 items

Package	Version	Source	Known Vulns	Notes
`openpyxl`	`*`	requirements.txt	No	Version not pinned — minor supply chain risk; recommend pinning to a specific version
`python-docx`	`*`	requirements.txt	No	Version not pinned — minor supply chain risk; recommend pinning to a specific version

Security Positives

✓ All network I/O is to documented, well-known endpoints (RSS feeds and the Volcengine ARK API)

✓ No sensitive paths accessed (no ~/.ssh, ~/.aws, .env, /etc/passwd, or similar)

✓ No credential harvesting — ARK_API_KEY is only used to authenticate to the Volcengine API for report generation

✓ No obfuscation, reverse shells, C2 communication, or data exfiltration

✓ No base64-encoded payloads executed in shell

✓ Dependencies are pinned (openpyxl, python-docx) and are standard, trusted packages

✓ No hidden instructions in comments or HTML

✓ Full environment variable set is passed to subprocess, not iterated for key harvesting

✓ Code is clean, readable, and purpose-built for the documented use case

✓ State deduplication prevents duplicate processing

✓ AI cache is cleared after each run (clear_summary_cache in finally block)

Scan Report

Findings 2 items

File Tree

Dependencies 2 items

Security Positives