Stop hoarding AI news. Turn the useful parts into tasks, gates, and local tools.
Most AI/news feeds are noise: model hype, screenshots, benchmark claims, safety incidents, and workflow ideas. News to Tools converts the parts worth acting on into concrete local state:
- workboard tasks with acceptance criteria and evidence
- guarded model candidates with
auto_route_allowed=falseuntil verified - usage/quota budgets for long-running agents
- medical AI claim gates that block clinical use without independent validation
- model security incident logs
- AI claim diligence for benchmark, cost, safety, deployment, adoption, and physical-AI claims
- PDF triage packets for long documents
- design handoff packets that turn product notes into component tasks and acceptance criteria
Most AI news creates a false sense of progress: you read more, save more, and build nothing. News to Tools is intentionally biased toward conversion:
- articles become implementation tasks
- benchmark claims become evidence-gated claim records
- model announcements become blocked candidates until verified
- product ideas route into existing repos before creating new ones
- risky medical/security claims stay behind explicit validation gates
It is for builders who want the useful part of a feed without letting hype silently change the system.
- Your reading queue is full but your shipped-tool queue is empty.
- You want model announcements captured as gated candidates, not automatically trusted dependencies.
- You want claims, incidents, PDFs, and article queues turned into local state with evidence requirements.
- You want a public, generic way to turn impressive AI claims into "implement", "track", "verify-first", or "reject" decisions.
python -m pip install -e .If your shell has not picked up the console-script path, use:
python -m news_to_tools --helpAdd an article-derived implementation task:
news-to-tools task-add "Implement article: desktop agent scheduler" \
--source "AI newsletter" \
--source-url "https://example.com/article"
news-to-tools task-listImport a generic queue JSON into the workboard:
news-to-tools queue-import examples/queue.sample.json
news-to-tools task-listBy default, queue-import imports items with missing/new/queued/todo/pending status and skips
done or dropped items. Use repeated --include-status STATUS flags when your queue uses custom
status names.
Queue import is idempotent. Re-running the same feed will not duplicate tasks; matching uses
source_url when present and falls back to the generated task title. Article priority and evidence
fields such as summary, notes, and source links are carried into the workboard.
Route a large article feed into product promotion targets:
news-to-tools promotion-plan examples/queue.sample.json
news-to-tools promotion-plan examples/queue.sample.json --format json --output promotion-plan.jsonThe planner groups news into existing products first, such as model routing, agent security, agent dogfooding, reusable skills, claim gates, and talent radar. It is meant to prevent duplicate repos from every hot article while still keeping the useful product signal.
Register a hyped model without letting your automation route to it:
news-to-tools model-add "example/SubQ-SSA" \
--source-url "https://example.com/model" \
--requirement independent_benchmark \
--notes "Claims are unverified; do not route production tasks."Block a medical AI claim until it has independent validation:
news-to-tools medical-claim "Model claims 91% doctor adoption" \
--source-url "https://example.com/medical-ai" \
--text "Claims 91% doctor adoption for medical record generation."Gate a broad AI claim before you build from it:
news-to-tools claim-add \
--subject "Example coding model" \
--domain coding \
--claim-type benchmark \
--claim "Solves more repository issues on a public benchmark." \
--source-url "https://example.com/coding-model" \
--benchmark "Public issue-resolution benchmark" \
--reproduction-evidence "local harness passed" \
--status validated
news-to-tools claim-list
news-to-tools claim-validateHigh-risk claims are intentionally stricter. Medical, security, and physical-AI deployment claims
need safety or reproduction evidence before they can move past verify-first.
Turn a design or product brief into buildable UI tasks:
printf "Search by zipcode and filter by price.\nExport selected results as a table.\n" > brief.txt
news-to-tools design-handoff brief.txt --add-to-workboard
news-to-tools task-listThe command writes a local handoff.json and handoff.md under .news-to-tools/design-handoffs/.
Each inferred component includes default/loading/empty/error states and acceptance criteria, then
can be imported into the workboard without duplicating tasks on rerun.
By default, the CLI writes local state under .news-to-tools/ in the current working directory.
That directory is ignored by git.
For automation, pin the state directory so commands do not depend on the current working directory:
news-to-tools --state-dir ./.news-to-tools-state task-list
NEWS_TO_TOOLS_STATE_DIR=./.news-to-tools-state \
news-to-tools task-add "Implement article: agent trace action extraction"This is not a scraper for private chats. The public repo intentionally excludes personal WeChat state, screenshots, and private queues. Bring your own article feed or queue fixture.
This is not a benchmark system. Model candidates are blocked from automatic routing until you record real local verification.
This is not medical advice. Medical AI claims are recorded as claims and blocked by default.