concinno-skills-content

Content-writer agent skills for Concinno — article extraction, Google News search, keyword extraction, rule-based text rewrite. Native Python only; no API keys needed for the MVP surface.

Status

MVP (0.1.0) — four tools, all offline-cleanable. This is the fourth sub-package in the concinno-skills-* ecosystem (after -google), built to the same entry-points + Tool-protocol pattern.

Tools

Install

pip install concinno-skills-content

Heads-up before first use:

• KeyBERT downloads a ~90MB SBERT model on first call (all-MiniLM-L6-v2 by default) to ~/.cache/huggingface/. One-shot per machine. Larger / multilingual models are selectable via the model kwarg; KeyBERT's own README lists the options. sentence-transformers itself is a hard dependency of keybert>=0.8 and will be installed automatically — the full dep chain including transformers + torch takes ~1-2GB of disk.
• newspaper3k + Python 3.12/3.13: newspaper3k has not seen an upstream release since 2018 and pins lxml. If pip install fails on 3.12+ due to a C-extension build, the usual fix is pip install --upgrade lxml_html_clean + the latest lxml wheel. Tracking issue: https://github.com/codelucas/newspaper/issues.

No OAuth, no API keys — every MVP tool uses public RSS / local compute only. _auth.py is a placeholder for future paid-API sub-packages.

Usage via Concinno ToolRegistry

When the consumer sets CONCINNO_LOAD_PLUGINS=1, the default registry auto-mounts all four tools:

import os
os.environ["CONCINNO_LOAD_PLUGINS"] = "1"

from concinno.tools.registry import get_default_registry

reg = get_default_registry()
names = set(reg.list_deferred())
assert {"ArticleExtract", "GoogleNewsSearch",
        "KeywordExtract", "TextRewrite"} <= names

tool = reg.get("ArticleExtract")
article = tool.call(
    action="parse",
    url="https://news.example/article-url",
    lang="en",
)

Direct Python usage

from concinno_skills_content import (
    ArticleExtract, GoogleNewsSearch,
    KeywordExtract, TextRewrite,
)

# 1) Discover
search = GoogleNewsSearch().call(
    action="search",
    query="AI agents",
    limit=5,
    lang="en",
    country="US",
    period="7d",
)

# 2) Extract
article = ArticleExtract().call(
    action="parse",
    url=search["items"][0]["url"],
    lang="auto",  # collapsed to "en" — newspaper3k has no real auto
)

# 3) Keyword SEO
kws = KeywordExtract().call(
    action="extract",
    text=article["text"],
    top_n=10,
    ngram_range=[1, 2],
)

# 4) Clean copy for downstream LLM
clean = TextRewrite().call(
    action="remove_ads", text=article["text"],
)["text"]
clean = TextRewrite().call(
    action="rewrite", text=clean,
)["text"]

All tools return either a structured success dict or {"error": "..."} — never raise. Matches the rest of the Concinno tool ecosystem.

License

Apache-2.0. See LICENSE in the Concinno monorepo.