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/malloydata/publisher/agents-mdgit clone --depth 1 https://github.com/malloydata/publisherWhat 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 | $0.05582 | $0.05582 |
| Opus 5 | $0.02791 | $0.02791 |
| Sonnet 5 | $0.01116 | $0.01116 |
| Haiku 4.5 | $0.00558 | $0.00558 |
Grade B, and why
publisher AGENTS.md scanned grade B with 2 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 yesterday.
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.
Reads agent configuration directoriesmediumAgent snooping
.claude/, .codex/, .gemini/ hold keys, settings and other credentials a mod has no legitimate need for.
Codex: add to `~/.codex/config.toml`: Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
curl -s http://localhost:4000/api/v0/status | jq .operationalState # -> "serving" How it starts
The opening of the file, as written. The whole thing — 194 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Working with Malloy Publisher
Publisher is the open-source semantic model server for Malloy. It serves one or more Malloy model packages over a REST API and a single MCP endpoint. If you are an AI agent working in this repo, here is what you can do with it and how to start.
What you can do
- Discover what data exists: environments, packages, models, sources, and fields, without knowing any names in advance.
- Answer plain-English questions by running Malloy queries, which Publisher compiles to SQL and runs against the connected database.
- Build and change Malloy models: validate an edit with
malloy_compile, save it, thenmalloy_reloadPackageto run it by name. Themalloy-modelingskill covers the workflow. - Build a data app: a hand-authored HTML page in a package's
public/directory, backed by that package's models and served by Publisher with no build step. Themalloy-html-data-appsskill covers it. - Review Malloy for correctness with the
malloy-reviewskill.
All of it runs against a local server you start in step 1 and reach over MCP in step 2, or over REST when you work unattended (section 7).
1. Start the server first
The MCP tools talk to a running server, so nothing works until it is up.
From a clone:
bun install
bun run build && bun run start # REST on :4000, MCP on :4040
To re-initialize the sample storage on a later run, build first and then start with --init: bun run build && bun run start:init. Without cloning, npx @malloy-publisher/server@latest --port 4000 runs the published build. Start one npx server at a time: concurrent first runs can race in the shared npx cache and corrupt the install (docs/deployment.md has the recovery step).
Keep the @latest. npx resolves through a shared cache and will happily re-run a build it downloaded weeks ago, so a bare npx @malloy-publisher/server can serve an old version while looking like a fresh start. The server does not report its own version, so a stale build is invisible until it behaves like one — a fixed bug that appears to still be there is the usual first sign.
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.
- yesterday First seen · 194 lines · 5,582 tokens per session scan B 69d4e5f6f333
publisher AGENTS.md is an instructions file published in the GitHub repository malloydata/publisher (99 stars, last pushed 2d ago), licensed MIT. It adds 5,582 tokens to every session, about $0.0279 per session on Opus 5. A static security scan graded it B with 2 findings (reads agent configuration directories, makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other instructions, from other repositories
evidence AGENTS.md
Instructions for evidence-dev/evidence, covering agents.md, this repo is a mirror, layout and scripts.
lightdash CLAUDE.md
Claude Code instructions for lightdash/lightdash, covering claude.md, agent skills, triage labels, domain docs and opt-in agent okteto development environment.
lightdash AGENTS.md
AGENTS.md instructions for lightdash/lightdash, covering agents.md, cursor cloud specific instructions, bringing the stack up (do this at the start of a session), process management and non-obvious caveats.
evidence CLAUDE.md
Instructions for evidence-dev/evidence, a project described as: Business intelligence as code: build fast, interactive data visualizations in SQL and markdown.
wren-engine AGENTS.md
Instructions for Canner/wren-engine: All project instructions live in .claude/CLAUDE.md. Please refer to that file for repository overview, build commands, architecture, and conventions.
spiceai copilot-instructions.md
Instructions for spiceai/spiceai, covering spice.ai agent instructions, data correctness — absolute top priority, evidence — no claim without a reproduction, build, test, lint (expensive — read first) and git & prs.