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/r0shi/harborclerk/agents-mdgit clone --depth 1 https://github.com/r0shi/harborclerkWrote 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/r0shi/harborclerk/agents-md)<a href="https://agentmods.dev/instructions/r0shi/harborclerk/agents-md"><img src="https://agentmods.dev/badge/instructions/r0shi/harborclerk/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 5.1 | $0.02154 | $0.02154 |
| Opus 5 | $0.01077 | $0.01077 |
| Sonnet 5 | $0.00431 | $0.00431 |
| Haiku 4.5 | $0.00215 | $0.00215 |
Grade A, and why
harborclerk 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 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 — 172 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AGENTS.md
Operating rules for any agent or human working in this repository. Rules here are binding. Reference material — architecture, API surface, schema — lives elsewhere; see Map. On matters of fact, the code is authoritative and this file may be stale: trust the source and fix this file. On matters of rule, this file is authoritative.
What this is
Harbor Clerk is a single-tenant, local-first document intelligence system: local extraction, OCR (eng/fra), hybrid retrieval with reranking, cited local-AI answers, and Research. It runs as a Mac-native app or via Docker Compose, and exposes the corpus to external agents through an MCP server and a mirrored CLI.
Non-negotiables
- Single-tenant. No
tenant_idin schema, API, or code. Ever. - Read-only external access. API keys are read-only and stored hashed.
- Snippets, not corpus. MCP/CLI callers receive retrieved passages and citations — never a bulk corpus export.
- Never modify a source. Watched-folder files are never altered, copied, or
moved. Mailboxes are read-only (IMAP
EXAMINE; read-only OAuth scopes when OAuth lands) — never marked read, never deleted. - Copy only where the source requires it. Watched-folder documents are read
in place from
source_path. IMAP is the deliberate exception: messages and attachments must be fetched and stored, because there is no file to reference. - All results carry citations. Every retrieval surface returns sources.
- Tika is required for PDF/Office/eBook/HTML/email extraction.
- Never commit credentials. Passwords, API keys, JWTs, or tokens supplied in conversation never enter a file, regardless of stated intent.
- Strategy stays private. Positioning, competitive analysis, claim posture,
and candid assessment live in
docs/strategy/(a separate private repo), never here.
Working agreement
- Branch first. Create a feature branch before any change, including docs.
Never commit to
main. Do this from the start, not retroactively. - PR always. Every change lands via PR, even docs-only. Branch protection is enforced.
- Never bypass the gate. No
gh pr merge --admin, no direct push to main, without explicit human permission in the current conversation. - Verify before claiming done. Run the checks and read the output before saying anything passes. Evidence, then assertion.
- Don't stack PRs. Land the base on
mainand branch again. A stacked branch once lost ~3,000 lines to a squash-merge of its base. - Regenerate docs after changing tools, routes, models, stages, or compose
services:
uv run python -m scripts.gen_docs. CI fails otherwise. dependency-auditcan fail on an unrelated PR. It audits the whole environment against live advisory databases, so a newly-published CVE turns it red on a change that touched nothing relevant.
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 · 172 lines · 2,154 tokens per session scan A a25705531397
harborclerk AGENTS.md is an instructions file published in the GitHub repository r0shi/harborclerk (2 stars, last pushed 2d ago), licensed MIT. It adds 2,154 tokens to every session, about $0.0108 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
pdfmux CLAUDE.md
Claude Code instructions for NameetP/pdfmux, covering pdfmux — agent instructions, project, commands, python and site.
memobase AGENTS.md
Instructions for maximskorohod/memobase, covering agents.md — установка плагина memobase, что это, способ b — через git (человек ставит вручную, без чата), безопасность and справочные факты (для сверки при установке).
docsift CLAUDE.md
Instructions for anishmoncivarghese/docsift, covering docsift, key documents (read before making product decisions), commands, architecture (src layout, package docsift) and hard rules.
skills CLAUDE.md
Claude Code instructions for linggen/skills, covering claude code instructions, edit workflow — three synced surfaces and conventions.
libratory AGENTS.md
AGENTS.md instructions for subev/libratory, covering libratory — agent context, product vision, task tracking, architecture and type safety — read this before writing code.
agentic-reviewer AGENTS.md
Instructions for FilippoScaramuzza/agentic-reviewer, covering agentic reviewer — journal paper review framework, project overview, directory structure, workflow and phase 0: setup & paper conversion.