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/jsharma1105/koshi/agents-mdgit clone --depth 1 https://github.com/jsharma1105/KoshiWrote 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/jsharma1105/koshi/agents-md)<a href="https://agentmods.dev/instructions/jsharma1105/koshi/agents-md"><img src="https://agentmods.dev/badge/instructions/jsharma1105/koshi/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 | $0.01869 | $0.01869 |
| Opus 5 | $0.00934 | $0.00934 |
| Sonnet 5 | $0.00374 | $0.00374 |
| Haiku 4.5 | $0.00187 | $0.00187 |
Grade A, and why
Koshi 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 — 173 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Koshi Sub-Agent Personas
Five focused personas that wrap the Koshi MCP server's 24 tools into role-specific agents you can invoke from MCP-compatible clients (GitHub Copilot CLI, Claude Code, Cursor, Windsurf, Agency, …).
The same five personas ship in two formats:
| Format | Location | Used by |
|---|---|---|
| Copilot CLI / Agency | .github/copilot/agents/*.agent.md |
GitHub Copilot CLI, Microsoft Agency CLI |
| Claude Code | .claude/agents/*.md |
Claude Code, Claude Desktop |
Personas
| Persona | Pillar | Tools | When to use |
|---|---|---|---|
koshi-librarian |
Retrieval | 5 + diagnostics | Index code/docs and search them. |
koshi-memory-keeper |
Memory | 9 + diagnostics | Store / recall facts, decisions, patterns, preferences; vault export/import/sync; turn-end auto-capture. |
koshi-context-packer |
Context | 3 + read-only retrieval/recall | Plan token budgets and pack prompt windows for cache reuse. |
koshi-quality-coach |
Quality | 5 + diagnostics | Score AI interactions per team, surface trends, recommend tuning. |
koshi-orchestrator |
All | All 24 | Generalist that routes cross-pillar requests. |
Each persona enforces single-responsibility boundaries — for example, koshi-librarian will refuse to call memory or team tools and will hand off to the right persona instead.
Prerequisites
Pick one — every flavor talks to the same MCP server.
Python (pip install koshi)
pip install koshi
That's it. The first Client() call auto-downloads a matching koshi-mcp
native AOT binary into your user cache, verifying it against the SHA-256
baked into the wheel. No .NET install required.
from koshi import Client
with Client() as koshi:
print(koshi.version())
The personas below are informational in Python — they describe how to scope the 24 tools into role-specific prompts when you wire Koshi into Claude Code / Copilot CLI from a Python-only host. The agent-installer below is .NET-only.
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 · 173 lines · 1,869 tokens per session scan A 9b7ee8a41c1e
Koshi AGENTS.md is an instructions file published in the GitHub repository jsharma1105/Koshi (2 stars, last pushed 13d ago), licensed MIT. It adds 1,869 tokens to every session, about $0.0093 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
headroom copilot-instructions.md
Copilot instructions for headroomlabs-ai/headroom: When performing a pull request review in this repository.
Acontext AGENTS.md
Instructions for memodb-io/Acontext, covering acontext, navigation guide, plan-driven development, modules and rules.
puppyone AGENTS.md
Instructions for puppyone-ai/puppyone, covering puppyone (contextbase), connect, collaborate, platform and active development directories.
framework AGENTS.md
AGENTS.md instructions for ai-driven-dev/framework, covering agents.md, behavior, communication, action and memory management.
robrain AGENTS.md
Instructions for adelinamart/robrain, covering robrain — context management (oss self-hosted), session start (mandatory, first thing in every new chat), after every response (mandatory) and session end (last thing).
unigent AGENTS.md
Instructions for gintasz/unigent, covering setup — do this first in any fresh checkout or worktree, core vs. adapter, constitution, validation and releases.