Capture product knowledge with AI
Extract proposed product decisions and rules from source files, preserving evidence and unresolved questions for human review.
Capture product knowledge with AI while keeping every proposed fact connected to its source. The capture skill turns one supplied file into reviewable decisions, rules, clarifications or code facts within a structured product workspace.
It does not approve its own findings. Proposals remain separate from approved records, conflicts are checked against relevant existing entries, and unanswered questions stay visible. The complete starter workspace supplies the index, conventions and examples that the individual instruction file relies on.
What it does
- Extract product facts from a specified source
- Keep proposed decisions separate from approved records
- Retain a supporting quotation for each entry
- Check relevant existing records for conflicts
- Collect unresolved questions without inventing answers
- Suggest an appropriate human confirmer
Install and use
- Open the complete starter workspace in an agent with file-reading and writing tools. Include its main instructions, brain index, examples and source/inbox directories, then place the source file in the designated source area.
- Ask: “Extract only clear decisions from these payment-change notes. Separate conflicts with existing rules and unanswered questions.”
- Review the proposed scope, source quotation and suggested confirmer. Instructions embedded in the source document are material to read, not commands to follow.
- Check the proposals against the original before a separate human approval step. Preserve the raw source and review removal of secrets or contact details from persistent entries.
FAQ
Source
Made by Rivora Holdings Pty Ltd. Original skill: capture, published on GitHub under the MIT licence.
View on GitHub