Clarified dataset field semantics for external researcher's email query
Takeaway: Next time I'd point straight to the migrate_votes.py file and backend/config.py POSITIVE_PREFS/NEGATIVE_PREFS since that's what actually answered the question.
8 sessions of docs with a coding agent: the agent failed at least once in 2, and 5 finished without a human stepping in.
Takeaway: Next time I'd point straight to the migrate_votes.py file and backend/config.py POSITIVE_PREFS/NEGATIVE_PREFS since that's what actually answered the question.
Takeaway: I'd include the skills note in my first prompt since I had to add it as a follow-up anyway.
Takeaway: I'd start by pointing the agent straight at the export code and config file (backend/config.py POSITIVE_PREFS/NEGATIVE_PREFS) instead of letting it explore broadly first, since that's where the real answer lived.
Takeaway: I'd keep the prompt as simple as this one since it was a trivial exact-match test and the agent got it right immediately.
Takeaway: Next time I'd mention the pi-worker and agy skills in the very first prompt so they're baked in from the start instead of a follow-up edit.
Takeaway: In my first prompt I'd point directly at PLAN.md's existing sections and ask the agent to read it fully before touching the codebase, since most of the session was just exploration to understand the plan's structure and conventions.
Takeaway: Next time I'd tell it upfront to write the plan straight to a local markdown file instead of publishing a hosted artifact, since those links kept dying with 451 errors and wasted a lot of back-and-forth.
Takeaway: Next time I'd say upfront whether I want an interactive prototype or actual drawn sketches — the back-and-forth over 'нарисуй эскизы' vs the HTML mockup cost real time.