comis CLAUDE.md

Project instructions for comisai/comis, an agent runtime that coordinates models, tools, memory, scheduling, approvals, and related operations. They tell coding agents how to preserve its generic design while making changes.

In plain words
What is it for?
Use them when changing comis code, especially runtime, prompts, configuration, or workspace starters. They guide where reusable mechanisms belong and where deployment-specific behavior should be placed.
Why use it?
They prevent application-specific rules or business knowledge from being built into the shared runtime. They also require agents to check the project's engineering guidance before editing code.

Instructions file

Install

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.

agentmods
npx agentmods add instructions/comisai/comis/claude-md
Clone the repo
git clone --depth 1 https://github.com/comisai/comis
Per session 7,394 This file is loaded in full into every session.
When invoked 7,394 The same file — it is already loaded in full.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.07394 $0.07394
Opus 5 $0.03697 $0.03697
Sonnet 5 $0.01479 $0.01479
Haiku 4.5 $0.00739 $0.00739

Measured 2d ago against content hash 7329fd0d62c1, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

comis 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.

CLAUDE.md · 234 lines

How it starts

The opening of the file, as written. The whole thing — 234 lines — stays where its author put it; the contents beside it link to each section on GitHub.

CLAUDE.md

Read AGENTS.md before any code change — it is the authoritative engineering protocol.

This file is a Claude-specific operational supplement. If anything here conflicts with AGENTS.md, follow AGENTS.md and update this file.

Generic Runtime Check — Before Every Code Change

Preserve Comis as a generic agent runtime on every change. Before editing production code, read docs/developer-guide/generic-agent-architecture.md and decide whether the requested behavior is a universal runtime mechanism or application-specific expertise.

  • Runtime code may provide reusable mechanisms for orchestration, models, tools, memory, channels, scheduling, approvals, delivery, observability, security, typed prompts, locale policy, and immutable workspace policy.
  • Do not add an industry, persona, business workflow, fixed human language, response script, domain vocabulary, or task-specific tool instructions to the engine prompt, workspace starters, default config, packages/core, or packages/agent.
  • Put deployment-specific policy and persona in operator workspace files. Put reusable procedures, examples, domain knowledge, and tool-selection guidance in an opt-in prompt skill. When it should ship with Comis, build a repository-shipped skill instead of specializing runtime code — AGENTS.md §6.5 owns which skill tree actually ships and seeds.
  • Put external API capabilities behind MCP or a typed adapter. Keep their instructions bounded and attributed, preserve side-effect/approval metadata, and never let integration or skill prose grant capabilities or override engine/operator policy.
  • Promote behavior into the runtime only when an unrelated deployment can use it without inheriting the requester's assumptions. Do not accept a temporary hard-coded special case.

During final diff review, ask: would a completely unrelated agent inherit any domain assumption from this change? If yes, move that behavior to workspace policy, a skill, MCP, or an adapter before calling the change complete. Run pnpm vitest run test/architecture/generic-runtime-boundary.test.ts for every change touching runtime prompts, workspace policy, locale behavior, integrations, health surfaces, or specialization boundaries; extend that gate when the new regression class is not already covered.

Read the full file on GitHub · 234 lines

Changes

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.

  1. 2d ago First seen · 234 lines · 7,394 tokens per session scan A 7329fd0d62c1

Subscribe to this mod's changes

comis CLAUDE.md is an instructions file published in the GitHub repository comisai/comis (5 stars, last pushed 2d ago), licensed Apache-2.0. It adds 7,394 tokens to every session, about $0.0370 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.