Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/richfrem/agent-plugins-skills/os-improvement-loopnpx skills add richfrem/agent-plugins-skills --skill os-improvement-loopgit clone --depth 1 https://github.com/richfrem/agent-plugins-skillsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/richfrem/agent-plugins-skills/os-improvement-loop)<a href="https://agentmods.dev/skills/richfrem/agent-plugins-skills/os-improvement-loop"><img src="https://agentmods.dev/badge/skills/richfrem/agent-plugins-skills/os-improvement-loop.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00074 | $0.00984 |
| Opus 5 | $0.00037 | $0.00492 |
| Sonnet 5 | $0.00015 | $0.00197 |
| Haiku 4.5 | $0.00007 | $0.00098 |
Grade A, and why
os-improvement-loop scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 5d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 84 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Concurrent Agent Loop
Treats concurrent sessions as OS threads sharing a common event bus and memory address space. Every cycle includes execution, independent peer evaluation, friction tracking, self-assessment surveys, post-run metrics, and memory persistence.
When to Use This Pattern
Use when:
- Coordinating continuous improvement across multiple concurrent agent sessions.
- Evaluating and improving multiple skills, workflows, or templates in parallel.
- You need every cycle to generate measurable accuracy gains and persistent memory.
Do NOT use for:
- Single-session procedural tasks (use
os-eval-runnerdirectly). - Signal-only coordination with no evaluation, survey, or memory steps.
Key Invariants
- No-Rollback Rule: Never manually roll back changes during a cycle unless
evaluate.pyregisters an explicit accuracy regression. - Eval Gate Mandatory: Every modification must pass the independent evaluation gate (
evaluate.pyexit code 0). No manual bypasses. - NEVER STOP Discipline: Do not abort a running loop due to minor/moderate errors. Complete the loop close checklist and log unresolved issues as Map Debt.
- Outer Loop Ownership: The outer loop owns session lifecycle. Inner loop tasks (
os-eval-runner) must not prematurely close a session without running Stage 4 (memory promotion and survey collection).
Stage Pointers & Reference Protocols
The execution details are split across modular references:
- Stage 0: Setup and Orientation — Pre-flight reads, registry, and packet design.
- Stage 1: INNER_AGENT Execution — Strategy execution, friction logging, and local scoring.
- Stage 2: PEER_AGENT Verification — Independent evaluation run and verdict formulation.
- Stage 3: Decision Logic — KEEP/DISCARD actions and correction packets.
- Stage 4: Loop Close Checklist — Surveys, ledger updates, memory promotion, and retrospectives.
- Orchestrator Meta-Survey — Meta-evaluation of loop coordination patterns.
- Lab Limits — Hard caps on iterations, eval sets, and critic invocations per run.
What ships with it
24 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
- assets/architecture-overview.md 57 B
- evals/evals.json 2.0 KB
- evals/results.tsv 73 B
- references/acceptance-criteria.md 42 B
- references/dual-loop.md 32 B
- references/lab-limits.md 33 B
- references/memory/improvement-ledger-spec.md 56 B
- references/memory/metrics.md 40 B
- references/memory/post_run_survey.md 48 B
- references/meta/backlog.md 38 B
- references/operations/triple-loop.md 48 B
- references/orchestrator-meta-survey.md 47 B
- references/program.md 30 B
- references/stage-0-orientation.md 42 B
- references/stage-1-execution.md 40 B
- references/stage-2-verification.md 43 B
- references/stage-3-decision.md 39 B
- references/stage-4-close.md 36 B
- references/testing/test-registry-protocol.md 56 B
- scripts/eval_runner.py 31 B runs code
- scripts/evaluate.py 28 B runs code
- scripts/generate_report.py 35 B runs code
- scripts/kernel.py 26 B runs code
- scripts/post_run_metrics.py 36 B runs code
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 5d ago First seen · 84 lines · 74 tokens per session scan A 4836abd9d6bc
os-improvement-loop is a skill published in the GitHub repository richfrem/agent-plugins-skills (6 stars, last pushed yesterday), licensed MIT. It adds 74 tokens to every session and 984 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
notebook
Project notes system to prevent AI context loss and reasoning loops. Init notes, save mid-flow, recover context, migrate messy notes, resolve lessons. Subcommands: /notebook, /notebook save, /notebook recover, /notebook migrate.
dotagents-standard
Set up, author, and navigate the dotagents standard — a slim AGENTS.md "router" at the repository root plus a hidden .agents/ directory (rules, context, memory, personas, skills, specs, logs, tasks) that splits agent context into small per-topic files loaded on demand (progressive disclosure). Use this WHENEVER the…
memento
Context handoff for fresh sessions. Saves unsaved session knowledge to notebook, then produces a short orientation block to paste into a new chat. Use when user says '/memento', 'hand off context', 'prepare for fresh chat', 'save and brief', or before clearing context. NOT for: recovering context (use /notebook…
sync-rules
Iteration closeout knowledge routing for AI-assisted development. Use when the user asks to summarize changed function points into a knowledge base, update project docs or generated superpower docs after code changes, sync rules, distill lessons from an iteration, record durable decisions, or remember user…
poly-wiki
多平台知识库编译器。从碎片素材中提炼结构化知识，维护可复用知识网络。 当用户要求编译素材、萃取知识、整理笔记、健康检查知识库、查询归档时触发。 Triggers: 知识库 wiki 编译 萃取 录入 整理 健康检查 lint query ingest extract 提炼知识 编译素材 整理笔记 知识管理 卡片笔记 Zettelkasten.
poly-wiki
Multi-platform knowledge base compiler. Compile raw materials into structured knowledge, maintain a reusable knowledge network. Triggers: wiki compile extract ingest lint query knowledge base notes Zettelkasten.