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 skills add krsnczky/agency-icm-builder --skill agency-icm-buildergit clone --depth 1 https://github.com/krsnczky/agency-icm-builderWrote 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/skills/krsnczky/agency-icm-builder/agency-icm-builder)<a href="https://agentmods.dev/skills/krsnczky/agency-icm-builder/agency-icm-builder"><img src="https://agentmods.dev/badge/skills/krsnczky/agency-icm-builder/agency-icm-builder/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/krsnczky/agency-icm-builder/agency-icm-builder"><img src="https://agentmods.dev/badge/skills/krsnczky/agency-icm-builder/agency-icm-builder.svg" alt="Reviewed on agentmods" width="80" 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 5.1 | $0.00135 | $0.02635 |
| Opus 5 | $0.00068 | $0.01318 |
| Sonnet 5 | $0.00027 | $0.00527 |
| Haiku 4.5 | $0.00014 | $0.00264 |
Grade A, and why
agency-icm-builder 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 12d 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 — 118 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Agency ICM Builder
Build workspaces where the folder structure runs the agency. One AI agent, reading the right files at the right moment, coordinates client work across services: the routing files decide what loads, per-client folders guarantee client data never mixes, and a session-end capture protocol turns daily work into durable memory.
This is not theory. The conventions here run a real agency: 15 client folders, 5 service departments, daily production use. Everything below was extracted from that system, then genericized.
The term ICM (Interpretable Context Methodology, folder structure as agent architecture) comes from Van Clief & McDermott, arXiv:2603.16021.
The ten rules
Every workspace this skill builds or audits follows these. In audit mode, each rule is a check.
- The root file routes, it does not teach. The root
CLAUDE.mdholds identity plus a routing table (task keyword to file path) and nothing else. Content lives in the files it points to. When the root file grows, it is absorbing payload that belongs on a shelf. - Every client is one folder. Nothing about a client lives outside
clients/<name>/. This is the no-mixing guarantee: the agent always knows which client it is working on, so client facts always land in the right folder. Cross-client insight goes to a shared patterns file, stripped of client identity. - Three files tell the client's story.
hot.md(current state and priorities, rewritten as things change),log.md(append-only event history with tags like[DECISION][RESULT][PROBLEM]),learnings.md(distilled operational rules). Now, what happened, what we learned. An agent reading these three files cold is up to speed. - Load per task, not per session. A load-order contract (one file in
system/) says which files each task type needs: a report task loads campaign state, a copy task loads brand voice, a strategy task loads the client profile. The agent loads those and stops. Whole-workspace loads are a bug. - One home per fact. A link beats a copy. Two hand-maintained copies of the same fact will drift, and drifted copies are worse than no copy, because one of them lies.
- New client = copy the template.
clients/_template/is the unit of instantiation. Evolution rule: when three real clients independently grow the same folder, the template gains it. One client's extra folder is an exception; three are structure. - State is rewritten, history is appended.
hot.mdgets edited in place.log.mdonly grows. Never rewrite history to reflect current state; that is whathot.mdis for. - Sessions end in capture. Work not written back to the client folder did not happen. The capture is agent-driven at session end: log entry, hot update if priorities shifted, learnings append if there was a lesson. If client attribution is uncertain, the note goes to a quarantine inbox for the human to file. Never guess the client.
- Expertise and state live apart.
departments/(orservices/) holds stable know-how: agent instructions and skills that change rarely and apply to every client.clients/holds per-client state.system/holds workspace infrastructure: policies, the load-order contract, changelogs. A fact that survives every client belongs in a department; a fact about one client never leaves its folder. - A hand-maintained index that duplicates state will rot: generate it, shrink it, or drop it. The failure mode is narrow: a file that re-lists state living elsewhere (client rosters, "map of everything" system maps) and relies on a human remembering to update it. That copy drifts and starts lying. Fix it one of three ways: rebuild it from a script, keep it small enough that drift is obvious at a glance (the root routing table is exactly this), or delete it if nothing reads it. Not every index is guilty: an index that holds its own content rather than duplicating state, and a purely presentational file (for example one that only exists to shape a graph view), are not state copies and this rule does not touch them. The question to ask is "does this file restate something whose real home is elsewhere?" Only then is it at risk.
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.
- 12d ago First seen · 118 lines · 135 tokens per session scan A 34e00a048ba0
agency-icm-builder is a skill published in the GitHub repository krsnczky/agency-icm-builder (11 stars, last pushed 1mo ago), licensed MIT. It adds 135 tokens to every session and 2,635 once invoked, about $0.0007 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 2026-08-30.
Other skills, from other repositories
systematic-debugging
Use when encountering any bug, test failure, or unexpected behavior, before proposing fixes.
local-ai-agents
Build local-first AI agents that run entirely on a developer workstation with Microsoft Foundry Local and Qwen function-calling models. Covers Small Language Models (SLMs), the OpenAI-compatible local endpoint, sandboxed local tools, local RAG with Chroma, local MCP servers, hybrid cloud/local routing, and the…
next-cache-components-adoption
Turn on Cache Components in a Next.js app and resolve the blocking routes it surfaces. Use when the user wants to enable, adopt, or migrate to Cache Components, flip the cacheComponents flag, work through a flood of blocking-prerender / instant validation errors, run the cache-components-instant-false codemod, or…
chat-pet-sprite-creation
Use when creating or changing VS Code chat pet sprite art, sprite sheets, state animations, eye treatments, Stable/Insiders variants, or pet transitions under src/vs/workbench/contrib/chat/browser/widget/media/chatPet.
cpu-profile-analysis
Analyze V8/Chrome CPU profiles (.cpuprofile) and DevTools trace files (Trace-.json). Use when: profiling performance, investigating slow functions, comparing code paths, finding bottlenecks, analyzing timeToRequest, understanding call trees from sampling profiler data, analyzing layout/paint/rendering, investigating…
insight-error-page
Write or audit an insight-kind error page for the Next.js dev overlay. Use when creating a new errors/ .mdx page, auditing an existing one, or checking that a page matches the framework fix cards. Covers page structure, title alignment, FixCard cards with Copy prompt button, code snippets, terminology verification…