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 instructions/tonydzi/charm-os/agents-mdgit clone --depth 1 https://github.com/tonydzi/charm-osWrote 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/instructions/tonydzi/charm-os/agents-md)<a href="https://agentmods.dev/instructions/tonydzi/charm-os/agents-md"><img src="https://agentmods.dev/badge/instructions/tonydzi/charm-os/agents-md.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 | $0.00895 | $0.00895 |
| Opus 5 | $0.00447 | $0.00447 |
| Sonnet 5 | $0.00179 | $0.00179 |
| Haiku 4.5 | $0.00089 | $0.00089 |
Grade A, and why
charm-os AGENTS.md 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 4d 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 — 71 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AGENTS.md — working in this repo
Written for AI coding agents, and equally readable by a human contributor. Short on purpose.
What this repo is
CharmOS — a framework that treats humans and AI agents as first-class relationships on top of a local-first markdown vault, driven by one loop: RDR (Recall → Deep Research → Synthesis).
It is manifest-first: the design documents are the product, and the modules are the parts of it
that are already real. Read MANIFESTO.md and docs/architecture.md
before proposing structure; a module that does not fit the RDR loop needs an argument, not a folder.
License: Apache-2.0 (not MIT like most of this lab's repos) — check the header expectations of any file you add.
Stack and layout
- Python 3, stdlib-first. Zero-network, zero-token where the module allows it.
modules/eval-harness/— the deterministic scorer.eval.pyloads a JSONL trace, groups events by proposal, runs every invariant ininvariants.py, prints a scorecard.sanitize.pyandcurate_public.pyproduce publishable fixtures from private traces.modules/rdr/rdr.py— the decision loop itself.modules/turnstate/— the per-turn memory ledger and its backfill.docs/— architecture, the RDR loop, privacy, and the category argument.examples/— synthetic vault and CRM entries. Every example is fictional. Keep it that way.
How to verify a change
cd modules/eval-harness
python eval.py benchmarks/consensus-safety-v0/fixture.jsonl
python eval.py benchmarks/public-live-v0/fixture.jsonl --json out.json
The exit code is the number of failed invariant checks (0 = clean), so CI can gate on it — do not "fix" a non-zero exit by changing the expectation. Same trace in, same score out: that determinism is what makes it a benchmark rather than a demo, and it is the property to protect.
Paste the scorecard in the PR. If you added an invariant, add the fixture line that fails it.
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.
- 4d ago First seen · 71 lines · 895 tokens per session scan A 83a7bc6266e8
charm-os AGENTS.md is an instructions file published in the GitHub repository tonydzi/charm-os (2 stars, last pushed 6d ago), licensed Apache-2.0. It adds 895 tokens to every session, about $0.0045 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 instructions, from other repositories
Kompl AGENTS.md
Instructions for tuirk/Kompl, covering agents.md — kompl operator guide, what is kompl?, agent rules — wiki safety (read before any docker command), never without explicit user confirmation and if a destructive action is going to happen anyway.
memex AGENTS.md
AGENTS.md instructions for memex-lab/memex, covering memex — project context for coding agent, build & run, testing requirements, architecture and layers.
memex CLAUDE.md
Claude Code instructions for memex-lab/memex, covering memex — project context for claude code, authoritative agent rules, product, core capabilities and target platforms.
memex copilot-instructions.md
Copilot instructions for memex-lab/memex: AGENTS.md is the canonical instruction file for this repository. Read it before suggesting or editing code.
memex GEMINI.md
Gemini CLI instructions for memex-lab/memex: AGENTS.md is the canonical instruction file for coding agents in this repository. Read it before editing code.
superbrain CLAUDE.md
Instructions for m3talux/superbrain, covering superbrain — repository rules, non-negotiables, public-facing content, versioning and workflow.