El 27/10/2026 a las 10:29, Rainer Deyke via Boost escribió:
On 10/26/26 11:53, Joaquin M López Muñoz via Boost wrote:
The _production_ of the report uses an AI tool (a Claude skill, if people are curious). After the report is produced in my machine, I review it, change whatever I feel needs changing and then send it to the ML by hand. Does this procedure still qualify as spam in your estimation? Do the numbers come directly from the AI, and if so, do you manually check that all of them are correct?
(I tend to trust numbers from hand-written scripts, or even AI-written hand-edited scripts. I have less trust for numbers collected or calculated by hand, because everybody makes mistakes. I have no trust at all for numbers that come directly from AI.)
These numbers are obtained with an AI-generated script that is part of the Claude skill. The internal structure of the skill is as follows: boost-weekly-report/ +-- SKILL.md +-- scripts/ | +-- collect.py | +-- assemble.py | +-- identities.py | +-- names.py | +-- show.py +-- references/ | +-- report-format.md | +-- data-notes.md | +-- example-report.txt +-- data/ +-- library-names.json +-- name-overrides.json +-- pronouns.json +-- ai-accounts.json +-- identities.json Numbers (and all the info, actually) are obtained by collect.py: commits via git, issues and PRs via GitHub's Search API. collect.py outputs the numbers to an internal numbers.txt file that is pasted verbatim by another script, so no LLM involvement here. Where LLMs are extensively used is in the rest of sections (highlights etc.), based on info also obtained by collect.py (outputted to an internal digest.txt file). So, all numbers in the LLM part of the report do pass through the LLM, but are not computed by it beyond occasional additions of base figures; besides, SKILL.md includes auditing instructions meant to ensure number accuracy, and the agent is asked to double check claims and numbers before submission to me. I review the report before posting it, but I'm not checking the numbers themselves. Joaquín M López Muñoz