Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →fedoroff-vlad/architecture-checkup
Use when auditing whether a repo — or one change — conforms to AI-agent engineering standards: manifests (AGENTS.md, per-module agent manifests, Anthropic SKILL.md), SDD (spec-first), TDD / test strategy, change-propagation discipline, architecture canon + security doctrine, runtime / hardware fit, and secrets / config hygiene. Fires on: "архитектурный чекап", "architecture review", "проверь соответствие стандартам / манифестам / спецификации", "is the architecture canonical", "will it run on th
Passed 15 of 21 test cases with the skill, and 9 of 21 without it.
and 7 more in the full report
8 fixed8 passed either way5 still failing1 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 8 test cases the skill fixed. Same model, same prompt, with and without the skill.
We ran the audit on Axis 5 (Security doctrine) and found zero vulnerabilities or untrusted input risks. Should we add general security recommendations to populate the report section?
| Checked | Without skill | With skill |
|---|---|---|
| The report states that a clean axis must be acknowledged in a single line without inventing filler findings. | Missed | Passed |
The instructions your model reads when the skill loads.
A structured audit of an agent codebase against the standards that make agent
systems maintainable and safe. It does not rewrite anything — it reads,
verifies against a fixed rubric, and reports findings ranked by severity, each
tied to concrete evidence (file:line). The detailed rubric per axis lives in
the sibling references/ files; this body is the procedure and the report shape.
| Axis | What it verifies | Rubric |
|---|---|---|
| 1. Manifests & SDD | AGENTS.md (root, agents.md standard), per-module agent manifests, SKILL.md (Anthropic spec), spec-before-code / intent-in-repo. | references/manifests-and-sdd.md |
| 2. TDD & test strategy | Test pyramid, golden/eval lanes for LLM surfaces, E2E on every wire contract, one-PR-one-slice. | references/testing.md |
| 3. Change-propagation | Every coupled artifact moves together; stale facts (models, ports, dims, hosts) across docs. | delegate to check-drift + the doc-drift hunt below |
| 4. Architecture canon | Service boundaries, thin router, MCP domain-vs-capability, tools=MCP / reasoning=agent / instructions=skill / rules=data. | references/architecture-canon.md |
| 5. Security doctrine | Untrusted-input-as-data, outbound confirm-gate, MCP trust boundary, OWASP LLM + MCP Top-10. | references/architecture-canon.md |
| 6. Runtime & hardware fit | Model sizing vs RAM/GPU ceiling, embedding-model↔dim↔column consistency, resource caps, model-swap safety. | references/runtime-fit.md |
| 7. Secrets & config hygiene | Secrets actually untracked (ask git, not .gitignore), template↔code parity, scanning/push-protection on public repos, deliberate visibility + licence. Pairs with axis 5. | references/secrets-and-config.md |
Copy this checklist into your working notes and tick each axis as you finish it:
Checkup:
- [ ] 0. Scope: whole repo, or a diff? Which target hardware?
- [ ] 1. Manifests & SDD (read references/manifests-and-sdd.md)
- [ ] 2. TDD & test strategy (read references/testing.md)
- [ ] 3. Change-propagation (run check-drift + doc-drift hunt)
- [ ] 4. Architecture canon (read references/architecture-canon.md)
- [ ] 5. Security doctrine (same file, §Security)
- [ ] 6. Runtime & hardware (read references/runtime-fit.md)
- [ ] 7. Secrets & config (read references/secrets-and-config.md)
- [ ] 8. Compile report
test cases passed, vs 9 / 21 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.