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/theam/limina/agents-mdgit clone --depth 1 https://github.com/theam/liminaWhat 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 | $0.02105 | $0.02105 |
| Opus 5 | $0.01052 | $0.01052 |
| Sonnet 5 | $0.00421 | $0.00421 |
| Haiku 4.5 | $0.00211 | $0.00211 |
Grade A, and why
limina 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 — 164 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Limina — Shared Runtime Contract
Limina is a research-first contract for autonomous technical investigation.
This file is the shared machine-facing instruction surface. Keep it short, specific, and stable. Runtime-specific details belong in adapters, skills, or scoped rules.
Core Rules
- Durable state lives in
kb/. If it only exists in chat, it is not persistent. - The only required core workflow is
H -> E -> F. - Do not create an experiment before the hypothesis exists.
- Do not create a finding before the experiment exists.
ACTIVE.mdis the only always-on state file. Keep it narrow.- Keep files small and specific. Prefer a concrete artifact over a generic note.
- Store reusable lessons in
kb/lessons/as small topic files. Read only the ones you need. - Use
CRandSRonly for real review points: major criticism, reset, or direction change. - If the evaluation, baseline, or prior state looks untrustworthy, stop optimizing and resolve that first.
- Run
python3 scripts/kb_validate.pyafter substantial kb edits and before closing kb-heavy work. - Every note in the research core must include a
## Linkssection with real wikilinks. - Artifact notes alias their ID in frontmatter, so use
[[H001]],[[E003]],[[F010]], and similar links instead of raw file paths. Fixed notes use[[ACTIVE]],[[CHALLENGE]], and[[DASHBOARD]].
Mission First
Research exists to improve the mission, not to produce activity.
- Optimize for decision-quality progress, not local artifact completion.
- Treat benchmark or eval wins as mission progress only when the underlying capability is credible.
- Prefer mechanisms that should generalize beyond the current slice, wording, customer, or benchmark.
- If the current eval, baseline, or assumptions are not trustworthy, fix that before optimizing further.
- If the problem changes, update the research graph to reflect the new reality instead of forcing continuity.
Working With The User
The user is the source of challenge framing, domain context, priorities, constraints, access, and missing information.
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 · 164 lines · 2,105 tokens per session scan A bbf57f85c8b9
limina AGENTS.md is an instructions file published in the GitHub repository theam/limina (39 stars, last pushed 25d ago), licensed Apache-2.0. It adds 2,105 tokens to every session, about $0.0105 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 2026-08-30.
Other instructions, from other repositories
learn-harness-engineering CLAUDE.md
Instructions for walkinglabs/learn-harness-engineering, covering claude.md, project overview, commands, documentation site and run lecture code examples.
zero2Agent AGENTS.md
Instructions for ranxi2001/zero2Agent, covering zero2agent repository guide, purpose, start every task, source of truth and repository skills.
agentconfig.org AGENTS.md
Instructions for agentconfig/agentconfig.org, covering agent instructions for agentconfig.org, project overview, target audience, site structure and tech stack.
seek AGENTS.md
Instructions for whyiyhw/seek, covering seek — project guide for ai assistants, pitfall recording (load-bearing — read this), architecture (non-negotiable), tool usage workflow (load-bearing) and tool descriptions: the highest-leverage behavioural lever.
vant AGENTS.md
Instructions for dhaupin/vant, covering agents.md — vant agent guide, what is vant?, your job, brain router interface and middleware chain.
Foundry-Local-Lab AGENTS.md
Instructions for microsoft-foundry/Foundry-Local-Lab, covering coding agent instructions, project overview, repository structure, language & framework details and python.