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/david-2814/metis/agents-mdgit clone --depth 1 https://github.com/david-2814/metisWrote 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/david-2814/metis/agents-md)<a href="https://agentmods.dev/instructions/david-2814/metis/agents-md"><img src="https://agentmods.dev/badge/instructions/david-2814/metis/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 | $0.27234 | $0.27234 |
| Opus 5 | $0.13617 | $0.13617 |
| Sonnet 5 | $0.05447 | $0.05447 |
| Haiku 4.5 | $0.02723 | $0.02723 |
Grade A, and why
metis AGENTS.md scanned grade A with 1 finding 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.
Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
- **Status-page live deployment recipe (Wave 15, 15a-4)** — [`docs/operations/status-page.md`](docs/operations/status-page.md) gains three new top-level sections: (1) "Live deployment" — honest about the hosting-account How it starts
The opening of the file, as written. The whole thing — 320 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AGENTS.md
Shared context for any AI agent working in this repo (Claude Code, Cursor, Codex, Aider, etc.). CLAUDE.md is a symlink to this file.
What Metis is
A local-first AI dev agent server. Provider-agnostic via a canonical message format, with a layered routing engine (manual / configured rules / learned patterns), bounded portable memory, and an event bus that feeds a trace store. Python core server; thin clients (CLI first, TUI/Tauri later).
Status: Phase 1 + Phase 2 + Phase 2.5 + Phase 3 shipped (owner-ratified 2026-05-16); Wave 16 reaches the GA launch milestone for the first paid cohort. The three Phase-3 wedges — transparent HTTP gateway, multi-user / per-team attribution, and evaluator — are live end-to-end with buyer-facing value. Wave 15 closed both pre-GA blockers (NETWORK escalation now requires repeated transient failures; SDK-canonical bare model names normalize before routing), shipped the Stripe-backed billing module, observability extensions, concierge tools, and the status-page deployment recipe. Wave 16 turns those primitives into launch operations: billing self-service portal + plan changes + failed-payment grace, first-customer concierge scaffolding + anonymized reports + case-study templates, and day-1 operational playbooks + status-page config artifacts. Wave 18 reaches the batch-API + compaction-substrate milestone: the Anthropic Batches API is wired end-to-end via the new adapter Protocol (submit_batch / poll_batch / fetch_batch) + canonical BatchHandle / BatchStatus / BatchError types + Usage.pricing_mode + ModelPricing.batch_rates, with both consumers shipped: metis evaluate --batch-mode (eval backfills) and scripts/benchmark.py --batch-mode (benchmark reruns). Live benchmarks/RESULTS.md §Wave-18 documents the exact 50% discount on fix-a-bug-small (sync $0.001220 → batch $0.000610 apples-to-apples). Compaction substrate ships alongside: CompactionCache SQLite store (per-workspace, content-hash keyed, LRU-evicted) + three session.compaction_* PSEUDONYMOUS events in the catalog. SessionManager wiring lands in Wave 19a-2; the substrate is dormant until then. Differentiator posture: delegation is the validated routing-surface GTM lever, with three cost-per-quality measurements at 8.3%, 19.9%, and 26.1% better than sonnet-only baseline on the fan-out workload. Slot-4 model selection remains a proof-of-mechanism, not a generalized regime: §A3-rev3 produced one sonnet inversion, while §A3-rev7 completion produced zero sonnet picks across 36 routing decisions and ruled out the last rubric/knob repair candidate on the dev-loop suite. The optional task-domain §A3 wedge is deferred post-GA. 1953 tests passing.
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 · 320 lines · 27,234 tokens per session scan A 82b7f87914d6
metis AGENTS.md is an instructions file published in the GitHub repository david-2814/metis (2 stars, last pushed 2mo ago), licensed Apache-2.0. It adds 27,234 tokens to every session, about $0.1362 per session on Opus 5. A static security scan graded it A with 1 finding (makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other instructions, from other repositories
kungfu copilot-instructions.md
Copilot instructions for kungfu-systems/kungfu: Read and follow AGENTS.md before proposing or running changes. It is the shared agent router for this repository, including the canonical Shifu development and build entrypoint.
GPT-RAG release.instructions.md
Instructions for Azure/GPT-RAG, a project described as: Enterprise-grade accelerator for agentic RAG on Azure. Built on Microsoft Foundry with Foundry IQ as the default retrieval backend, Microsoft Agent Framework orchestration, Zero-Trust architecture and IaC.
aeon CLAUDE.md
Instructions for aeonfun/aeon, covering aeon, how aeon works, strategy, voice and soul file hierarchy (read in this order).
nanobot CLAUDE.md
Claude Code instructions for HKUDS/nanobot, a project described as: Ultra-lightweight, open-source, self-hosted personal AI agent framework in Python with WebUI, tools, memory, MCP, multi-agent workflows, automation, and chat apps.
agentic-harness AGENTS.md
AGENTS.md instructions for frangelbarrera/agentic-harness, covering project context, architecture in 30 seconds, non-negotiable rules (from manifesto.md), coding standards and commit conventions.
initrunner CLAUDE.md
Instructions for vladkesler/initrunner, covering initrunner and project map.