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 iwe-org/dev-workspace --skill setupgit clone --depth 1 https://github.com/iwe-org/dev-workspaceWrote 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/iwe-org/dev-workspace/setup)<a href="https://agentmods.dev/skills/iwe-org/dev-workspace/setup"><img src="https://agentmods.dev/badge/skills/iwe-org/dev-workspace/setup/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/iwe-org/dev-workspace/setup"><img src="https://agentmods.dev/badge/skills/iwe-org/dev-workspace/setup.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 | $0.00079 | $0.00938 |
| Opus 5.5 | $0.00032 | $0.00375 |
| Sonnet 5.5 | $0.00016 | $0.00188 |
| Haiku 4.5 | $0.00008 | $0.00094 |
Grade A, and why
setup 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 1mo 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 — 68 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Workspace setup
Turn the blank workspace into this project's memory. The codebase is the primary source — read it first and draft from evidence; ask the developer only what the code can't answer. Every future session reads what you write here.
Steps
- Assess. Read
data/product.md. List which ✏️ blocks are unfilled. If everything is filled, ask what to revise instead of re-onboarding. - Locate the codebase. Ask where the project's code lives (often the parent or a sibling directory of this workspace, or this workspace may sit inside the repo itself). If the project is greenfield — no code yet — skip to step 4 and run the interview alone.
- Scan. Read the repository's README, package manifests, build
configuration, entry points, directory layout, and test setup. Draft from
what you find:
data/product.md— What is it, Platforms, Stack from direct evidence; leave ✏️ plus a concrete question under any section the code can't answer (Users, Constraints usually need the developer).- One starting
data/architecture/<slug>.mddescribing the module layout and any design decisions visible in the code (state management, storage, process boundaries). Say "unknown" where you'd be guessing. (The deep, per-module map underdata/codebase/is out of scope for setup — file it as follow-up work rather than attempting it here.) - Propose (don't yet write) spec docs for the 2–3 most load-bearing behaviors you can identify.
- Interview. Ask in batches, conversationally:
- Product: the one-liner; who uses it and for what; who it's not for.
- Reality: current stage, what's shipped vs. aspirational, the next thing they intend to build.
- Constraints: performance budgets, compatibility promises, licensing, privacy — anything every plan must respect.
- Rules: recurring instructions they find themselves repeating to agents or contributors — these become the Authoring rules section.
- Write. Fill every
data/product.mdsection, deleting the italic instruction lines as sections fill; add a dated entry to its Changelog. Write the confirmed architecture doc(s) and any spec stubs the developer approved, linking each fromdata/architecture.md/data/spec.md. - Close the loop. Mark the finished onboarding tasks
(
fill-product-doc, andcapture-current-architectureif step 3 ran):iwe update -k data/backlog/<slug> --set stage=done --set completed=<today>, and move their links indata/backlog.mdto## Done. Delete the example docs — every*.example.mdunderdata/(iwe delete <key>per doc).iwe deleteremoves their inclusion links from hubs automatically but flattens inline links to plain text: sweep the hub files and remaining docs for leftover de-linked example lines and remove them. - Validate & commit.
iwe normalize, theniwe schema validate— both must pass clean. Commit with a message likesetup: product doc filled, architecture captured, examples removed.
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.
- 1mo ago First seen · 68 lines · 79 tokens per session scan A 144651a61948
setup is a skill published in the GitHub repository iwe-org/dev-workspace (10 stars, last pushed 2mo ago), licensed MIT. It adds 79 tokens to every session and 938 once invoked, about $0.0003 per session on Opus 5.5. 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 skills, from other repositories
agent-state-bar
A design guide for an agent status bar: a short, automatically updated summary of progress, remaining tasks, tool counts, errors, and system state placed near the latest model input.
context-engineering
A guide to organizing the instructions and information an AI agent receives at each step. It covers system prompts, tool descriptions, conversation history, on-demand skills, and protections against unsafe outside content.
kv-cache-design
A guide to arranging an AI agent's conversation so previously processed text can be reused. A KV cache stores intermediate results for earlier tokens, while a prompt cache reuses a stable beginning across requests.
loop-engineering
A guide to building reliable loops for AI agents that plan, act, check their work, and continue until a task is truly complete. It includes independent verification and a proposer-reviewer pattern in which one agent's work is checked using outside evidence.
multi-agent-design
A guide to deciding when several AI agents should work together and how they should share tasks, context, and files. It compares peer, manager-led, and decentralised arrangements and explains common coordination failures.
reward-design
A guide to defining the scores that an AI agent receives during trial-and-error training. It covers checks based on rules, human preferences, or another model, and distinguishes rewarding the final result from rewarding the steps taken.