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/dmora/agentrun/claude-mdgit clone --depth 1 https://github.com/dmora/agentrunWhat 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 | $0.05375 | $0.05375 |
| Opus 5 | $0.02687 | $0.02687 |
| Sonnet 5 | $0.01075 | $0.01075 |
| Haiku 4.5 | $0.00537 | $0.00537 |
Grade A, and why
agentrun CLAUDE.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 — 190 lines — stays where its author put it; the contents beside it link to each section on GitHub.
agentrun — Claude Code Project Guide
Library-First Mindset
agentrun is a public Go library — not an application. Every decision must prioritize external consumers:
- Composability: interfaces should be wrappable, decoratable, and mixable without friction. Consumers build their own orchestrators on top of agentrun primitives.
- Extensibility: adding a custom backend (CLI or API) should require implementing 1–2 small interfaces, not understanding the whole codebase. No closed registries or internal-only extension points.
- Greenfield: no backwards compatibility concerns. Design the best API possible without legacy shims or deprecation paths.
- Think like a library author: exported API surface is a contract. Keep it small, intentional, and hard to misuse. Unexported internals can change freely.
Design Philosophy — Root is Language, Backends are Dialect
See DESIGN.md for full rationale, examples, and anti-patterns.
The root package defines the shared vocabulary for all backends:
- Output vocabulary:
MessageTypeconstants (what agents say) - Input vocabulary:
Option*constants (what you ask of agents) - Structural config:
Session.Model,Session.Prompt
Backend packages translate vocabulary into their wire format (CLI flags, API bodies).
Decision rule for where a constant lives:
Would this concept exist if backend X didn't exist? Yes → root. No → backend package.
Examples: OptionSystemPrompt → root (every LLM has one). OptionPermissionMode → claude/ (Claude CLI sandboxing).
Anti-pattern: Don't place cross-cutting constants in a backend just because only one backend exists today. Design for the intended architecture (N backends), not the current snapshot.
Build & Test Commands
make qa # full quality gate (tidy-check + lint + test-race + vet + vulncheck + examples)
make check # fast check: lint + test (no race detector)
make test # go test -count=1 ./...
make test-race # go test -race -count=1 ./...
make lint # golangci-lint run ./...
make vet # go vet ./...
make cover # test with race + coverage report
make tidy-check # verify go.mod/go.sum are clean
make vulncheck # govulncheck ./...
make bench # benchmarks with memory allocation stats
make fuzz # fuzz tests (30s per target)
make fmt # gofmt -w .
make tidy # go mod tidy
make examples-build # cd examples && go build ./...
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 · 190 lines · 5,375 tokens per session scan A aadcd8733d4a
agentrun CLAUDE.md is an instructions file published in the GitHub repository dmora/agentrun (5 stars, last pushed 7d ago), licensed MIT. It adds 5,375 tokens to every session, about $0.0269 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
workflow AGENTS.md
AGENTS.md instructions for vercel/workflow, covering agent instructions, architecture, core components, workflow execution model and development commands.
NEEDLE CLAUDE.md
Instructions for jedarden/NEEDLE, covering needle project conventions, msrv, module dependency graph, code style and testing.
a-i--skills 03-development-workflow.instructions.md
Instructions for organvm-iv-taxis/a-i--skills, covering development workflow instructions, no repo-wide build system, common development commands, after adding/modifying skills and validation before committing.
a-i--skills 02-skill-format.instructions.md
Instructions for organvm-iv-taxis/a-i--skills, covering skill format instructions, skill.md file structure, required yaml frontmatter, optional yaml fields and naming rules.
MindFlock AGENTS.md
Instructions for MindFlock/MindFlock, covering mindflock — agent onboarding, setup and tests, map and conventions.
agentis AGENTS.md
Instructions for agentis-labs/agentis, covering agentis - agent setup instructions, quick start, get an api key, bootstrap rules and reflect an existing setup.