Dev-Mem AI Programming Experience Knowledge Base

Problem

Debugging traces, pitfalls, version constraints, and fixes often remain trapped in temporary chat context. Once a session ends, that knowledge is easy to lose; the next time you hit a similar error, API limit, or design tradeoff, you may have to investigate again. dev-mem turns useful engineering experience from the current conversation into a local, searchable knowledge base instead of letting it scroll out of the window.

How It Works

It is triggered by @emem or natural language and routes to different workflows: quick capture, enrichment, search, daily distillation, session review, statistics, recent records, version checks, knowledge base repair, backup, and feature pre-checks.

During capture, it extracts structured fields from context, such as symptom, constraints, investigation chain, root cause, fix, conclusion, and caveats, without requiring the user to restate everything. It then creates or appends to dev-mem.md, generates IDs like EXP-CODE-202604-003, and keeps category statistics updated.

It supports mark anchors so review can prioritize flagged nodes; a diagnostic-tree mode for one symptom with multiple root causes; project-name inference; fuzzy search; and pre-check hints. Feature pre-checks only trigger with @emem plus forward-looking intent, surfacing historically relevant pitfalls with relevance levels.

Boundaries

dev-mem can only read the current context window and cannot recover early content from prior sessions; long conversations may lose earlier context, so timely capture or session-end review is safer.

Pre-check quality depends on accumulated entries; low hit counts do not imply low risk. Project-name inference has platform limits, and a Knot workspace root should not be treated as a project name. Repair workflows fix formatting rather than content, so back up before larger structural changes.