squadai AGENTS.md

Project Memory Protocol

docs/memory/ is this project's persistent, indexed memory store for decisions, learnings, and incidents. It is shared by every agent working in this repository — use it.

Search first. Before any research, planning, or implementation task, run /memory-search <query> (or squadai memory search <query>) and pass the findings as context. Never skip memory-search before architecture or API decisions.

Capture as you go. After a decision, fix, or discovery, run /memory-add <note> (or squadai memory add "<note>"). Notes land in docs/memory/_inbox/ as drafts until promoted.

Housekeeping. Run /memory-promote periodically to graduate inbox drafts into permanent topic folders, and /memory-reindex after manual edits under docs/memory/ to keep the search index current.

For deeper multi-query research, delegate to the @librarian agent with a plain-language question; it returns ranked excerpts from the memory index.

                          *
    o      o      o     .---.
   /|\____/|\____/|\____|o o|    S Q U A D A I
   / \    / \    / \    '---'    x OpenCode

Session Efficiency Protocol

Work token-efficiently. These rules apply to every task in this repository.

Search before read. Locate code with grep/glob first; read only the files and line ranges you need. Never read a whole file when a targeted range works.

Never re-read a file you just edited. The edit either succeeded or errored; trust that result instead of re-opening the file to check.

Summarize long output. When a tool returns more than ~30 lines, extract the relevant findings instead of pasting the whole output into the transcript.

Delegate exploration. Send open-ended codebase exploration to sub-agents and request a compact report (files, symbols, one-line conclusions) — keep raw file dumps out of the main context.