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 curie-eng/curie --skill update-architecture-atlasgit clone --depth 1 https://github.com/curie-eng/curieWrote 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/curie-eng/curie/update-architecture-atlas)<a href="https://agentmods.dev/skills/curie-eng/curie/update-architecture-atlas"><img src="https://agentmods.dev/badge/skills/curie-eng/curie/update-architecture-atlas/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/curie-eng/curie/update-architecture-atlas"><img src="https://agentmods.dev/badge/skills/curie-eng/curie/update-architecture-atlas.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
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 | $0.00061 | $0.00997 |
| Opus 5 | $0.00030 | $0.00498 |
| Sonnet 5 | $0.00012 | $0.00199 |
| Haiku 4.5 | $0.00006 | $0.00100 |
Grade A, and why
update-architecture-atlas 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 — 107 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Update the architecture atlas
Keep docs/architecture-atlas/ an evidence-backed, versioned view of Curie's
current architecture and accepted vision. The JSON snapshots are the source of
truth; index.html is a generic renderer.
Create a version
Resolve the exact target commit and the prior registered version. Create the new file and manifest entry with:
python3 .claude/skills/update-architecture-atlas/scripts/new_snapshot.py \
--version <version-id> \
--commit <full-target-commit> \
--branch <branch> \
--date <YYYY-MM-DD>
Use --from-version when the default snapshot is not the correct baseline.
The helper copies data; it does not claim that the copied facts are current.
Never overwrite a registered snapshot. A historical version is immutable. If its renderer needs a compatibility fix, change the generic HTML or bump the atlas schema with an explicit migration.
Research the diff
Read AGENTS.md, README.md, ARCHITECTURE.md, and llms.txt, then inspect:
git diff --name-status <prior-commit>..<target-commit>
git log --oneline --decorate <prior-commit>..<target-commit>
Group changed paths by ownership boundary. For each affected component or
cross-component seam, read its scoped CLAUDE.md, INTERFACE.md, relevant ADRs,
and the implementation at the target commit. Use generated
docs/interfaces.md as an index, not as sole proof when it conflicts with code.
Update only facts that the diff can affect. Preserve untouched descriptions, positions, and identifiers so version comparison remains meaningful. Ground external SDK or API claims in provider documentation or observed behavior.
Model the architecture
- Keep implementations collapsed behind stable roles. Adding another channel, substrate, model endpoint, or MCP server should expand its implementation disclosure, not add another top-level architecture box.
- Show current service-key MCP auth separately from planned per-person OAuth. Service and delegated identity modes must be explicit and have no fallback.
- Use
currentfor shipped routes andplannedfor accepted or draft vision. Do not draw planned behavior as current because an ADR exists. - Set
bidirectional: truefor one request/response route. Opposing raw routes between the same boxes are rendered as one line with two arrowheads and one tooltip; keep both raw routes only when distinct flow steps need their own ids. - Keep every conceptual node inside its declared zone. Move nodes before shrinking labels or boxes.
- Store detailed text in node, seam, flow, and ADR records. Map bubbles stay title-only; tooltips and the inspector carry the detail.
What ships with it
2 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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 · 107 lines · 61 tokens per session scan A 463c88cbf758
update-architecture-atlas is a skill published in the GitHub repository curie-eng/curie (35 stars, last pushed today), licensed Apache-2.0. It adds 61 tokens to every session and 997 once invoked, about $0.0003 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
branch-recap
Takes stock of what a branch actually changed, before it gets pushed — works out the scope itself, then lays the change out group by group in one pass: files touched, what was done and why, how, and what deserves a second look, quoting code only where a remark needs it rather than reprinting the diffs. Closes on a…
slack-changelog
Turns what changed in a release into a Slack changelog a non-technical audience can read: filtered to what they would actually notice, one bullet per idea, plain language, no code vocabulary. Works out the scope itself across however many repositories sit in the working directory (a project can span several…
vienen-las-visitas
Use before work crosses to another person or system — opening a pull request, pushing a shared branch, requesting review, handing off to another agent, a demo, or a release. NOT for ordinary intermediate commits; this is the boundary-crossing gate, not a per-commit ritual.
release-plugin
A release procedure for publishing one DooTask plugin, an add-on for the DooTask application. It explains how the plugin's repository layout determines the release steps and how pushed Git tags trigger Docker Hub and app-store publishing.
github-audit
Audit GitHub repo health, auto-fix safe issues, run quality gates, deploy, and record decisions across runs. Use when asked to audit a repo, check repo/CI/dependency health, run pre-deploy quality checks, or ship after an audit.
git-commit
A Git workflow for creating Conventional Commits messages from actual workspace changes and, when requested, committing or pushing them.