Extracted from scroll git history.
- DEC-001: Build scroll as incremental system with state tracking for reliable knowledge extraction (high)
-
- Large repositories would be expensive to re-process entirely each time
-
- API rate limits and costs make incremental processing essential
-
- Error recovery needed - partial completion should preserve progress
-
- DEC-002: Put knowledge where agents are forced to look rather than requiring tool usage (high)
- Don't ask agents to use a tool - put the knowledge where they're already forced to look. This guarantees knowledge delivery without requiring agent cooperation. Auto-sync via git hooks keeps knowledge fresh automatically.
- DEC-003: Bridge scroll to engram with deposit module for knowledge system interoperability (high)
-
- Enables knowledge flow from automated extraction (scroll) to manual curation (engram)
-
- Preserves source provenance with scroll-extracted tag and source field
-
- Prevents duplicates through state tracking
-
- LRN-001: Commit messages are thin - PRs and issues contain the real knowledge (high)
- Commit messages are brief and lack context. PRs and issues contain the real knowledge: reasoning behind decisions, alternatives that were debated, root causes of problems, and detailed discussions. Review comments often contain rejected alternatives that are valuable to capture.
- LRN-002: Knowledge systems need health scoring and staleness detection for maintenance (high)
- Knowledge systems accumulate cruft over time. Automated quality assessment helps maintain institutional memory by flagging entries that need review, identifying gaps in coverage, and detecting duplicates that dilute signal.