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/cirwel/unitares/claude-mdgit clone --depth 1 https://github.com/cirwel/unitaresWrote 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/cirwel/unitares/claude-md)<a href="https://agentmods.dev/instructions/cirwel/unitares/claude-md"><img src="https://agentmods.dev/badge/instructions/cirwel/unitares/claude-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 | $0.06204 | $0.06204 |
| Opus 5 | $0.03102 | $0.03102 |
| Sonnet 5 | $0.01241 | $0.01241 |
| Haiku 4.5 | $0.00620 | $0.00620 |
Grade B, and why
unitares CLAUDE.md scanned grade B 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 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.
- **User-level `~/.claude/hooks/`**, wired in `~/.claude/settings.json` (auto-test, watcher-hook, watcher-chime, stop-milestone-audit, …). Watcher findings at session start or in a chime block come from this chain, not f How it starts
The opening of the file, as written. The whole thing — 316 lines — stays where its author put it; the contents beside it link to each section on GitHub.
CLAUDE.md — unitares
Bootstrap for Claude Code sessions in this repo. Content below the SHARED CONTRACT markers is kept byte-identical with AGENTS.md — CI (scripts/dev/check-shared-contract.sh) enforces parity. Edit shared rules in both files; edit the Claude preamble here only.
The installable Codex/Claude adapter bundle is canonical in the companion unitares-governance-plugin repo. This file only governs how Claude Code should behave while working directly inside the unitares server repo.
Claude-specific wiring
Claude Code runs through a plugin-style harness. The hook lifecycle is owned by the unitares-governance-plugin repo (canonical for the adapter bundle); this repo vendors no hook chain of its own. When CLAUDE.md or AGENTS.md describes hook behavior, the plugin is the source of truth.
Three layers fire on Claude lifecycle events, and it matters which one produced what you are looking at:
- Plugin hooks —
hooks/session-start,hooks/post-edit,hooks/post-stop. The per-turn check-in is the plugin'spost-stop, which submitsepistemic_class="substrate_interpretation"for both Claude and Codex: a substrate reading of turn shape, not an agent-authored report. - User-level
~/.claude/hooks/, wired in~/.claude/settings.json(auto-test, watcher-hook, watcher-chime, stop-milestone-audit, …). Watcher findings at session start or in a chime block come from this chain, not from the plugin'spost-edit. - A file on disk under
hooks/is not evidence that it runs. Retired hooks stay on disk; checksettings.jsonbefore assuming a user-level hook is live.
To close a Watcher finding (the agent itself lives in this repo):
python3 agents/watcher/agent.py --resolve <fingerprint> # confirmed bug, fixed
python3 agents/watcher/agent.py --dismiss <fingerprint> # false positive
Reference fingerprints in the commit message — Watcher's audit trail lives in commits, not in tracked files (data/watcher/ is gitignored).
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 Changed · -17 lines · -713 tokens per session 5b05958a502a
- 5d ago First seen · 333 lines · 6,917 tokens per session scan B e166b3b3fa50
unitares CLAUDE.md is an instructions file published in the GitHub repository cirwel/unitares (4 stars, last pushed today), licensed Apache-2.0. It adds 6,204 tokens to every session, about $0.0310 per session on Opus 5. A static security scan graded it B with 1 finding (reads agent configuration directories). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other instructions, from other repositories
mempalace CLAUDE.md
Claude Code instructions for MemPalace/mempalace, covering claude.md, the mission, design principles, contributing and setup.
genlayer-project-boilerplate CLAUDE.md
Instructions for genlayerlabs/genlayer-project-boilerplate, covering claude.md, quick commands, linting, testing and deployment.
MemMachine AGENTS.md
AGENTS.md instructions for MemMachine/MemMachine, covering memmachine agent guide, repository orientation, environment setup, build, lint, and test commands and python (repo root).
Test Generation
Comprehensive Pytest conventions, mocking strategies, and model usage rules for OpenWebUI plugins.
streamlit agents.instructions.md
Instructions for streamlit/streamlit, covering agent definitions, file format, best practices, cross-platform compatibility and slash command support.
Cortex CLAUDE.md
Claude Code instructions for cdeust/Cortex, covering cortex — persistent memory mcp server, problem statement, build & test, releasing and architecture.