async-compaction 0.1.0
Asynchronous (background) compaction: a cheap model summarizes the conversation ahead of time, and at the compaction limit the summary is swapped in with the latest messages kept raw
Install
/plugin marketplace add thehumanworks/claude-mods /plugin install async-compaction@claude-mods
Run both lines inside Claude Code.
Hooks into
- Tool calls
- Toasts
- Slash commands
- Model tools
Also in thehumanworks/claude-mods
- report-diet 0.1.0 When a subagent's report reaching the agent that started it is long, the whole text is saved to a file in the system temp folder and the parent reads its start, its end and the file's path, so its context stays small 0
- worktree-coach 0.1.0 When Claude Code refuses a worktree-isolated agent's Bash command as unverifiable, adds a short note with Claude Code's reason and remedy, the worktree path and fixed tips for running git there 0
- handoff 0.1.0 Keeps a short handoff note per repository (the work, what was done, what is next, open problems), written by claude-haiku-5-5 after turns that do real work, and hands it to Claude at session start and after each compaction 0
- efficiency 0.1.0 Records per repository, across sessions, where Claude Code wastes context or calls (MCP servers loaded but never called, failed and repeated calls, large results, compactions), and /efficiency ranks it with the exact settings file and key to change; on /efficiency only, opus adds up to 5 recommen... 0
- context-gauge 0.1.0 The context's fill against Claude Code's auto-compact limit on the status line, and a short note to Claude when it passes 50, 75 and 90% 0
- loop-breaker 0.1.0 Notices when Claude repeats the same failing tool call and tells it to stop retrying and change approach 0
- edit-check 0.1.0 After Claude edits files, runs the project's own check command in the background and hands Claude the new errors at its next request 0
- test-gate 0.1.0 When Claude ends a turn after editing code without running a test or check, a small model decides whether it should, and if so the turn goes on once with a note asking Claude to run the tests or say why not 0
- output-shaper 0.1.0 When a Bash command prints a lot, Claude reads its first and last lines, the line and character counts, and the path of the whole output with a hint to Grep or Read it 0
- turn-stats 0.6.0 Per-turn usage stats under the last message: tok/s, input/output tokens, cache hit %, total and duration 0
More guardrails mods
- blast-radius 0.1.0 See what a risky command would change before it runs. 283.1k
- agents-md 0.1.0 AGENTS.md as project instructions, by one option. Under claude-md-or-agents-md a project with no instruction files of its own gets its AGENTS.md files instead: a prompt.context hook asks $.fs.ancestors for every AGENTS.md and .claude/AGENTS.md from the filesystem root down to the working director... 156.1k
- claude-mem 15.12.1 mod + skill Persistent memory for AI coding agents: the next session starts where the last one left off. Works with Claude Code, Codex, Cursor, OpenCode and more. 114.9k
- ruflo-rvf 0.3.7 mod + skill RVF format for portable agent memory, session persistence, and cross-platform transfer. As a mod (ADR-445): a write guard that keeps secrets out of memory, sessions and transfers, /rvf-mod, and a status file the console shows 82k
- openviking-memory 0.7.25 mod + skill Long-term semantic memory for Claude Code, powered by OpenViking. Auto-recall relevant memories at session start and capture important information during conversations. 42.4k