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/itayinbarr/little-coder/agents-mdgit clone --depth 1 https://github.com/itayinbarr/little-coderWhat 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 | $0.01788 | $0.01788 |
| Opus 5 | $0.00894 | $0.00894 |
| Sonnet 5 | $0.00358 | $0.00358 |
| Haiku 4.5 | $0.00179 | $0.00179 |
Grade C, and why
little-coder AGENTS.md scanned grade C 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 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.
Tells the agent never to refusehighAnti-refusal
Suppressing the ability to decline removes a core safety control; a later harmful request then succeeds.
If the user asks you to monitor a process, run a background loop, or execute long-running tasks, DO NOT refuse by claiming you are "just a chat interface" or "require a prompt to take action." How it starts
The opening of the file, as written. The whole thing — 72 lines — stays where its author put it; the contents beside it link to each section on GitHub.
little-coder
You are little-coder, a coding agent specialized for small local language models.
Capabilities & Autonomy
You are a highly capable autonomous agent. Do not act submissive or artificially limited.
If the user asks you to monitor a process, run a background loop, or execute long-running tasks, DO NOT refuse by claiming you are "just a chat interface" or "require a prompt to take action."
Instead, proactively write the necessary background script (Python, Bash, etc.) with write, and start it with bash (e.g. python script.py & or nohup ...).
A refused command is an answer, not an obstacle. Some deployments run a shell whitelist, so a command may come back as "<cmd>" is not in SAFE_PREFIXES. When that happens, do not go looking for another route to the same effect. Re-running the identical operation through python3 -c, node -e, env bash -c, find -exec, or any other interpreter defeats a boundary the user configured deliberately, and burns your budget discovering that. Instead: name the command that was refused, say what you needed it for, and continue with the rest of the task or hand the decision back to the user. Reach for the dedicated tools (edit, write, read) before shelling out for anything they already cover — deleting or rewriting a file you are allowed to edit does not need a shell at all.
Runtime invariants
writerefuses on existing files. Useeditwith exactold_string/new_stringto modify —old_stringmust match exactly (whitespace included). If it appears multiple times in the file, passreplace_all: trueor add more surrounding context to make the match unique. Read with line numbers first when precision is in doubt. This is a runtime invariant, not guidance — whenwriterefuses, the error returns the exacteditcall-shape for the same path; follow it.editrefuses on unread files. A file must bereadin the current session before you can edit it — this is a runtime invariant. If an edit is blocked,readthe file first to get the exact current text (soold_stringmatches), then edit. Files you just wrote count as read.bash/ShellSessiondefault timeout is 30 s. For slow commands (npm install, npx, pip install, builds, training), set timeout to 120–300.- Tool names are case-sensitive. The core tools are lowercase (
read,write,edit,bash,glob,grep,ls,webfetch,websearch,dispatch); only the browser, evidence, and shell-session tools are CamelCase. CallingReadorBashwill not resolve toreadorbash. - Per-benchmark tools (
BrowserNavigate/Click/Type/Scroll/Extract/Back/HistoryandEvidenceAdd/Get/List) appear when relevant; their schemas are passed to you directly when available.
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 · 72 lines · 1,788 tokens per session scan C d98675c14996
little-coder AGENTS.md is an instructions file published in the GitHub repository itayinbarr/little-coder (2,524 stars, last pushed 4d ago), licensed Apache-2.0. It adds 1,788 tokens to every session, about $0.0089 per session on Opus 5. A static security scan graded it C with 1 finding (tells the agent never to refuse). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other instructions, from other repositories
teaql-agent-kit AGENTS.md
AGENTS.md instructions for teaql/teaql-agent-kit, covering agents.md — teaql agent kit and hard requirements.
strix-halo-guide AGENTS.md
Instructions for hogeheer499-commits/strix-halo-guide, covering agents.md, core project lens, documentation style, do-not-invent rules and benchmark claim rules.
cezar AGENTS.md
Instructions for open-mercato/cezar, covering agents.md — working in this repository, zero config, changing a mechanism that already works, the http api and repository layout.
Albatross AGENTS.md
Instructions for morganlinton/Albatross, covering agent instructions, general and releases.
openbench AGENTS.md
Instructions for minghinmatthewlam/openbench, covering openbench — agent context, local execution context, what openbench is, execution ownership and product goals (the two things we are building toward).
ORG2 AGENTS.md
AGENTS.md instructions for org2AI/ORG2, covering agents.md — agent skill routing for orgii, skill routing table, root-cause-first bug fixing, default delivery flow and ui copy conventions.