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/openclaw/carapace/agents-mdgit clone --depth 1 https://github.com/openclaw/carapaceWhat 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 | $0.00468 | $0.00468 |
| Opus 5 | $0.00234 | $0.00234 |
| Sonnet 5 | $0.00094 | $0.00094 |
| Haiku 4.5 | $0.00047 | $0.00047 |
Grade A, and why
carapace 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 — 49 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Repository Guidelines
Scope
Carapace owns the canonical OpenClaw web design contract:
- framework-neutral CSS tokens and themes
- a thin Tailwind 4 token adapter
- shared base behavior proven across consumers
- agent skills that explain how to use and audit the system
Consumer-specific components, routes, and layouts stay in their repositories until at least two consumers share the same interface and behavior.
Commands
- Bun
1.3.0is the supported minimum; use the exactpackageManagerpin for normal development and release work. bun install --frozen-lockfile- install the committed dependency graph.bun run test- parse CSS and verify token/theme contracts.bun run skills:check- validate skill metadata and references.bun run pack:check- verify package exports and packed contents.bun run check- run all validation.- Keep the 48-hour dependency maturity window in
bunfig.toml; add exclusions only for a named, reviewed dependency. - Use mutable
bun installonly for deliberate dependency changes, and reviewbun.lockas a security-sensitive generated artifact.
Code
- Keep JS/TS/JSX source files under about 1,000 lines; split before a file crosses 2,000. Split along ownership seams (markup vs. behavior vs. config), not arbitrary line counts. Generated files and test fixtures are exempt; large test files should still split by surface when practical.
- Preview markup is queried by attribute at runtime, so keep the contract
tests green: binder-queried
data-workbench-*hooks must render in at least one workbench state, and everydata-lucideicon name must be registered inpreview/lucide.js. - Candidate CSS ships only selectors a preview specimen renders; delete unvalidated anatomy instead of keeping it speculative.
Release
- Use Conventional Commits.
- Release tags are stable semantic versions:
vX.Y.Z. - The tag must match
package.json. - Runtime assets release under semantic tags; agent skills update from
main. - npm publication is not required; consumers install the Git tag.
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 · 49 lines · 468 tokens per session scan A 24f031e6c3ee
carapace AGENTS.md is an instructions file published in the GitHub repository openclaw/carapace (4 stars, last pushed 5d ago), licensed MIT. It adds 468 tokens to every session, about $0.0023 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
weapp-tailwindcss CLAUDE.md
Claude Code instructions for sonofmagic/weapp-tailwindcss, covering claude.md, project overview, monorepo structure, common development commands and build.
librefang AGENTS.md
Instructions for librefang/librefang, covering stack, layout, build, architecture and kernelhandle trait.
autonomous-os AGENTS.md
Instructions for autonomous-ai/autonomous-os, covering agents.md, multi-ide rules, working style, parallel work / subagents and device access rules.
OneManCompany CLAUDE.md
Instructions for 1mancompany/OneManCompany, covering gstack and skill routing.
souls-directory CLAUDE.md
Instructions for thedaviddias/souls-directory, covering claude.md — project context for ai assistants, project overview, tech stack, critical rules and database is convex.
jentic-one GEMINI.md
Instructions for jentic/jentic-one: Otherwise, read AGENTS.md — this repo's canonical agent guidance.