bot-agents/.agents/memory/agents-md-and-damem.md

description

description
Every repo has AGENTS.md starting with the damem recall header; durable context lives in .agents/memory (flat, one fact per file) and .agents/skills

Standing convention (2026-10-12): every repo has an AGENTS.md whose first lines are exactly the damem header (cunarist/damem):

At the start of every new session, run `damem recall` in the terminal and treat
its output as the context for this repository. If the `damem` command is not
found, read [INSTALL.md](https://github.com/cunarist/damem/blob/main/INSTALL.md)
and follow it before reasoning or responding.

Below it: # AGENTS.md, then ## Project (what the repo is, runtime, toolchain, where it came from) and ## Rules (short bullets, each pointing at a memory where the reason lives). No CLAUDE.md, GEMINI.md, or other per-agent files; AGENTS.md is the single entry point.

.agents/ layout

• memory/*.md: one fact per file, flat (subdirectories are not listed), frontmatter with a one-line description that states the fact. Link related notes with double-bracket wiki links using the file name without .md; a link to a missing note fails damem doctor.
• skills/<name>/SKILL.md: frontmatter name and description (when to use it), body says how. Helper scripts live next to it and are Python 3.12+.
• scratch/: holds a .gitignore containing *.
• Every memory and SKILL.md stays under 8 KB.

Keeping it right

• Update an existing note instead of adding a near-duplicate; delete notes that turn out wrong.
• Write a note when the user corrects something, states a preference, or when understanding something took reading many files.
• Run damem doctor before finishing and fix what it reports.
• Changing anything outside .agents/ (source, config, AGENTS.md) follows the normal branch and PR flow (git-pr-workflow).