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/nagi-inaba/pmgs-reference/agents-mdgit clone --depth 1 https://github.com/Nagi-Inaba/pmgs-referenceWhat 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 | $0.01441 | $0.01441 |
| Opus 5 | $0.00720 | $0.00720 |
| Sonnet 5 | $0.00288 | $0.00288 |
| Haiku 4.5 | $0.00144 | $0.00144 |
Grade A, and why
pmgs-reference 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 2d 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 — 135 lines — stays where its author put it; the contents beside it link to each section on GitHub.
PMGS Reference Agent Guide
Purpose and users
This repository converts a locally acquired JPO PMGS package into one versioned SQLite source of truth and several read-only reference surfaces.
The local surfaces serve Python, CLI, and stdio MCP users.
The public surfaces serve people and AI clients that need source-backed patent-classification definitions through lightweight HTML, Markdown, JSON, and OpenAPI.
Do not turn this project into a patent-classification recommender, legal-opinion system, or general AI analysis service.
Language
Japanese is the canonical and default user-facing language.
- Keep
README.md, CLI help, the distributed skill, and the public root andllms.txtin Japanese by default. - Keep English available through
README.en.md, matching English guides,language=en,/en/, andllms.en.txt. - Update Japanese and English counterparts together when a shared public contract changes.
- Do not translate or rewrite official source text unless the output labels it as derived material.
Read before changing
Read only the documents relevant to the requested change, starting with:
PLAN.mdfor v1 scope, exclusions, architecture, and acceptance criteria.docs/current-status.mdfor measured state, unresolved external checks, and the next action.docs/requirements-traceability.mdfor requirement status and completion evidence.docs/architecture.mdfor component and data boundaries.- The applicable ADR under
docs/decisions/before changing an established decision. docs/release-runbook.mdanddocs/github-publication-checklist.mdfor release-related work.
Update the applicable source-of-truth document with any behavior, requirement, architecture, release-state, or operational change.
Do not report a local build as deployed, published, indexed, or externally available.
Repository ownership map
src/pmgs_reference/owns inventory, parsing, normalization, SQLite schema, queries, CLI, stdio MCP, and deterministic public exports.worker/owns route resolution, content negotiation, input validation, security headers, R2 reads, and optional WebMCP registration.schemas/owns JSON contracts and cross-language normalization vectors.config/publication-policy.yamlowns the fail-closed public-data boundary.tests/fixtures/synthetic_pmgs/is the only PMGS-shaped package allowed in tests and CI.docs/evidence/contains preserved public evidence and extracted text; treat source evidence as read-only unless an evidence refresh is explicitly requested.scripts/verify_repository_boundary.pyenforces the tracked and untracked candidate-file publication boundary.
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.
- 2d ago First seen · 135 lines · 1,441 tokens per session scan A 4489cf3d3144
pmgs-reference AGENTS.md is an instructions file published in the GitHub repository Nagi-Inaba/pmgs-reference (1 stars, last pushed 8d ago), licensed Apache-2.0. It adds 1,441 tokens to every session, about $0.0072 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
baerly-storage CLAUDE.md
Toolchain, verification matrix, module map, conventions, anti-patterns. The main agent entry point.
unsurf AGENTS.md
Instructions for acoyfellow/unsurf, covering agents.md — unsurf, what this is, stack, architecture and directory.
context CLAUDE.md
Claude Code instructions for Supa-Media/context, covering agent instructions — context, non-negotiables, 1. the customer owns the storage, 2. tenancy is bucket-level, never prefix-level and 3. plain files stay canonical.
stoop CLAUDE.md
Instructions for mislavjc/stoop, covering stoop agent notes, workspace layout, commands, effect conventions and schema contracts.
baerly-storage AGENTS.md
Instructions for Gusto/baerly-storage, a project described as: Document database with no database server. Tiny, LLM-friendly TypeScript API; DB engine runs to completion in each JS request handler; state lives in S3/R2.
skills AGENTS.md
Instructions for qdrant/skills, covering qdrant skills, project overview, project structure, navigating skills locally and conventions.