Watch
0
0
Fork
You've already forked hyperhive
0
hyperhive/claude-plugins/plugins/base/skills/state-hygiene/SKILL.md
atlas a429b30448 rename claude.md to agents.md for agents and the repo
Claude Code (>= 2.1.277) and opencode both read AGENTS.md, but Claude
Code reads it only when no CLAUDE.md exists in its cwd or above. So
this is a plain rename with no CLAUDE.md stub left behind.

- hive-agent/src/agents_md_migrate.rs: new. At harness start, before
  the first turn, renames state/CLAUDE.md to state/AGENTS.md when only
  the former exists; logs an error naming both paths when both exist
  and touches neither; does nothing otherwise. Idempotent.
- hive-agent/src/main.rs: calls it in serve_main; checkpoint prompt
  and a comment name AGENTS.md.
- hive-agent/src/claude_md_watch.rs: the size watch reads AGENTS.md;
  todo text, doc comments and tests follow. Module name, todo
  subsystem key ("claude_md") and metric name
  (hyperhive.agent.claude_md.lines) are unchanged so existing todo
  rows and dashboard history carry over.
- hive-agent/src/turn.rs, hive-agent/prompts/system.md: prompts tell
  the agent its file is AGENTS.md.
- hive-agent/src/{otel_turn_metrics,stats}.rs,
  frontend/packages/agent/src/stats.js, hive-subagent-mcp/src/*.rs:
  comments and a tool parameter description naming the agent's file.
- nix/host-modules/swarm-grafana/dashboards/agents.json: panel title
  and description.
- claude-plugins skills claude-subagents, state-hygiene: they mean the
  agent's own state file.
- CLAUDE.md -> AGENTS.md at the repo root; docs/crates/README.md link
  (checked by check-doc-refs.sh), hive-c0re/README.md,
  .prettierignore and the nix/sources.nix comment follow.
- docs/: every prose mention of the agent's file; persistence.md
  documents the boot-time rename, claude-invocation.md the shadowing.

Fixes #5509
2026-10-18 02:33:06 +02:00

3 KiB

name description
state-hygiene Enforces safe handling of an agent's own durable state/notes files (e.g. state/notes.md, state/TODO.md, or anything else under state/ meant to survive a container restart) - read the file before archiving, overwriting, or pruning it, split resolved work into small dated archive files instead of growing one giant live file, and never shell-pipe-append (>>) into a notes file. Use this whenever you are about to write to a durable notes/state file, especially right before a context-compaction checkpoint, when told to "flush state" or "write down anything relevant", or when a live notes file has grown large and needs an archive pass.

State Hygiene

Durable notes/state files (your AGENTS.md, state/notes.md, state/TODO.md, and anything else under state/) are your memory across container resets and context compactions. They only work if you actually read them before you change them - writing blind is how history gets silently destroyed or duplicated.

Before writing to any durable notes file

  1. Read it first. Never overwrite or append to a notes/state file based on memory or assumption of its current contents - the file may have moved on since your last read (another turn, another agent touched shared space, a prior session ended mid-edit). Read (or re-read the relevant section) immediately before editing.
  2. Never pipe-append. Don't do echo "..." >> state/notes.md (or equivalent) from a shell command. That's a blind write with no chance to see - and reconcile with - what's already there. Use your file-edit tool (read, then edit/write) instead.
  3. Prefer a targeted edit over a full rewrite for anything long-lived: an edit is less likely to accidentally clobber unrelated content than regenerating the whole file from your current context.

Keeping a notes file lean

  • Split into a live file (only what's currently active/gated/relevant - this is what you re-read every turn, keep it short) and an archive subdirectory (e.g. state/notes-archive/) for resolved/closed work, moved out of the hot path, never deleted. Don't drop archived files loose at the top level of state/ alongside the live files - a dedicated subdir keeps ls state/ showing only what's live.
  • Archive in dated, per-topic chunks - YYYY-MM-DD-<topic>.md - not one ever-growing archive blob. ls on the archive subdir should read like a timeline.
  • When you finish a piece of work, collapse its entry in the live file to a one-line "done" summary and move the full detail into a new dated archive file.
  • A live file well past ~500 lines (or into four figures) is overdue for an archive pass - do it at a natural checkpoint (finishing a task, a context-compaction prompt, or whenever you notice it).

If your environment ships its own hive-wide/repo-wide notes convention doc, treat that as the canonical source and this skill as a reminder to follow it - don't duplicate its rationale into your own notes either.