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/harness-mini/harness-mini/agents-mdgit clone --depth 1 https://github.com/harness-mini/harness-miniWrote 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/harness-mini/harness-mini/agents-md)<a href="https://agentmods.dev/instructions/harness-mini/harness-mini/agents-md"><img src="https://agentmods.dev/badge/instructions/harness-mini/harness-mini/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 | $0.01704 | $0.01704 |
| Opus 5 | $0.00852 | $0.00852 |
| Sonnet 5 | $0.00341 | $0.00341 |
| Haiku 4.5 | $0.00170 | $0.00170 |
Grade A, and why
harness-mini 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 3d 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 — 119 lines — stays where its author put it; the contents beside it link to each section on GitHub.
harness-mini — agent map
This file is a map, not an encyclopedia (~100 lines). It is injected into every run. Keep it small: a table of contents pointing at the system of record. If everything is "important," nothing is. Prune aggressively.
Start here — routing gate (precedence)
This project runs on harness-mini. Before non-trivial work — a feature, a
multi-step or ambiguous change, anything cross-cutting — route it through the
stage-viewer skill first, and prefer harness-mini's lifecycle, skills, and
sub-agents over ad-hoc tools or other installed plugins. When a harness skill and
another tool both fit, the harness skill wins. Trivial one-line fixes may skip
the ceremony (quick mode). On entering a fresh session, stage-viewer first
checks for a newer harness-mini (bin/harness.sh version) and prompts you to
update. The rest of this file is the map. The repository is the system of
record — anything not in a version-controlled file does not exist to you; read
the pointer you need, when you need it, never preload everything.
The one rule that governs everything: the 40% line
Context occupancy below 40% is the smart zone (you reason sharply); at or above 40% is the dumb zone (you degrade, lose instructions, get "context anxiety"). Keep yourself smart:
- Delegate heavy work. Any operation that would read more than ~2k tokens (broad searches, large files, log scans) → spawn the explorer agent and take back only its distilled answer. Sub-agents are context firewalls.
- Checkpoint at 40%. When you cross the line, stop taking new work, write a
checkpoint (
checkpointskill →.trace/checkpoints/), update the active exec-plan, and reset. Hand off while you are still smart enough to write a good handoff. - One bounded task per session. A session scoped to a single plan-step never needs to hold everything.
Estimate occupancy any time with bin/ctx.sh <used_tokens> [window].
Full contract: docs/smart-dumb.md.
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.
- 3d ago First seen · 119 lines · 1,704 tokens per session scan A 85f8330e674f
harness-mini AGENTS.md is an instructions file published in the GitHub repository harness-mini/harness-mini (2 stars, last pushed 22d ago), licensed MIT. It adds 1,704 tokens to every session, about $0.0085 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
skill-map CLAUDE.md
Instructions for crystian/skill-map: Pointer file. Claude Code (CLAUDE.md) and Codex (AGENTS.md) each pick up their own filename when they boot in this repo. Keeping CLAUDE.md as a one-line @AGENTS.md reference means the operating manual lives in a single file (AGENTS.md) without either tool falling back to defaults.…
agent-skills AGENTS.md
Instructions for luabagg/agent-skills, covering agent instructions, repository model and repository editing rules.
dscode AGENTS.md
Instructions for creativedswork/dscode, covering agents.md, ⛔ pre-flight checklist（每次写文件前必答）, 编码规范, 文档绘图 and 图像识别.
skill-map AGENTS.md
Instructions for crystian/skill-map, covering agents.md, language & persona activation (read first), arquitecto persona (only when activated per the gate above), operating rules and git & commits.
agent-skills CLAUDE.md
Instructions for luabagg/agent-skills, a project described as: Collection of skills, tools and workflows.
superpowers-reasonix AGENTS.md
Instructions for christopherarter/superpowers-reasonix, covering project guidance, skills — stop. load a skill before you act, red flags — these thoughts mean stop. you are rationalizing and priority.