architecture-review
Skill1 starsChanged 5 months agoMITView on GitHub
A periodic audit of the WHOLE codebase against the project's own conventions and hard constraints, not a review of the current session's diff. Surfaces convention drift that accumulated in code nobody touched recently, and reports it as a table you triage. Run every few sessions as hygiene, not every edit.
84 other projects use a version of this.
Adopted by
Most recently taken up by
- seancabalse/agentic-playbooktheir own version, first seen Nov 16, 2026
- Sequela02/canastatheir own version, first seen Nov 15, 2026
- OmarAlex24/audit-skillstheir own version, first seen Nov 15, 2026
- chuanglibs/Claude-Code-Game-Studios-Codextheir own version, first seen Nov 15, 2026
- myshra777-ai/garudatheir own version, first seen Nov 15, 2026
- Donchitos/Claude-Code-Game-Studiostheir own version, first seen Nov 9, 2026
Used with
Projects with it also use create-stories, story-done and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
lilangMaxSkillcreate-stories
Break a single epic into implementable story files. Reads the epic, its GDD, governing ADRs, and control manifest. Each story embeds its GDD requirement TR-ID, ADR guidance, acceptance criteria, story type, and test evidence path. Run after /create-epics for each epic.
Also in 11 projects that use this
lilangMaxSkillstory-done
End-of-story completion review. Reads the story file, verifies each acceptance criterion against the implementation, checks for GDD/ADR deviations, prompts code review, updates story status to Complete, and surfaces the next ready story from the sprint.
Also in 11 projects that use this
- +10 more from lilangMax
Other versions84 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 84
- Donchitos/Claude-Code-Game-Studios★ 39ksame name
- langwatch/langwatch★ 5ksame name
- ghuntley/loom★ 1.7ksame name
- microsoft/hve-core★ 1.6ksame name
- DataDog/dd-trace-js★ 840same name
- decebals/claude-code-java★ 763same name
- borgbase/vykar★ 722same name
- appdevforall/CodeOnTheGo★ 586same name
- owainlewis/blueprint★ 432same name
- getsentry/warden★ 424same name
74 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository3 other files it gives its agents.
- guardSkill
- handoff-updateSkill
- refactorSkill
The file
We split this skill into the rules it gives an agent
8 rules. 0 are common, 7 are distinctive, and 1 is very specific to this project.
0
Common
Most of the 76 edited versions we read keep these.
7
Distinctive
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 39
If you script the mechanical passes, the script is an accelerator, not a replacement.
- Distinctive: line 44
Run the mechanical collector if you have one, or the greps by hand, to get the raw hits.
- Distinctive: line 47
Run the judgment-only checks the collector can't: cohesion, duplication, derived-fact-in-two-places.
- Distinctive: line 64
Every few sessions, as a standing hygiene check.
- Distinctive: line 65
After any session that adds new infrastructure: new modules, new data flows, new config.
- Distinctive: line 66
After extracting a new shared/canonical helper: confirm nothing else still re-implements the same concern.
- Distinctive: line 67
Before a milestone, as a final pass.
- Very specific: line 69
Not every session, and not on the working-tree diff.
“Common” is measured: kept by at least 6 in 10 of the 76 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Architecture Review
The hard constraints in the agent contract are only as good as your ability to find where they're violated.
This skill audits the whole codebase against them on a cadence and reports the violations as a table. It is
the codebase-level counterpart to the guard skill: guard checks each action while you work, this catches what
slipped past, accumulated, or was never caught in the first place.
Not Claude-specific. Packaged as a skill, but it's a checklist run on a schedule. Adapt the invocation; keep the discipline. The checks themselves are your conventions: seed them from your contract's hard constraints, one check per rule.
This is NOT a diff review: audit the whole codebase
The single most important rule. A violation is a violation regardless of when, or by which session, it was introduced. "Pre-existing, out of scope" is never a reason to skip, downgrade, or omit a finding. It is the exact failure this skill exists to prevent. The default instinct, for an agent and a tired human alike, is to scope a review to what was just touched. But convention drift accumulates precisely in the code nobody touched recently, and only a whole-codebase pass surfaces it. If you find yourself writing "pre-existing," stop: report it as a row. The one acceptable carve-out is a finding already tracked as an open backlog item, and even that is reported as a row noting its tracked status, not silently dropped.
Two kinds of check
Each convention falls into one of two buckets, and they need different handling:
- Mechanical checks: grep-able patterns: banned names (
utils/helpers/misc), a forbidden call shape, a module missing its docstring, a bare catch-all exception. These can be scripted into a collector that emits raw hits, as a per-project accelerator. The collector never classifies: its grep nets are broad and will match deliberate, sanctioned sites; the triage is this skill's job, not the script's. - Judgment checks: the ones a grep can't do: module cohesion (does a file mix unrelated concerns), the same fact derived in two places, a decision that should be computed but is left to the model, a duplicated function body. These need reasoning over the code, not a pattern match.
If you script the mechanical passes, the script is an accelerator, not a replacement. When it's missing or errors, fall back to running the greps by hand. The convention list stays the authoritative definition.
Protocol
- Run the mechanical collector if you have one, or the greps by hand, to get the raw hits.
- Triage every hit: real violation, or a sanctioned site (a deliberate canonical home, a documented false positive, an already-tracked item). The judgment is the point; the grep only narrows where to look.
- Run the judgment-only checks the collector can't: cohesion, duplication, derived-fact-in-two-places.
- Report everything in one table. Don't skip a check; if it found nothing, note that in the summary line, not as a row.
Output format
| # | Location | Finding | Convention | Fix | Effort |
Then a summary line: "N findings across C checks. M quick fixes (< 10 min), K need tracking as backlog items."
Hand-off. Cohesion findings (a file mixing unrelated concerns, not "a file over N lines") chain straight
to the refactor skill. Everything else is user-directed: present the table and let the operator pick what to
act on. The audit's job is to find and report, not to silently start fixing.
When to run
- Every few sessions, as a standing hygiene check.
- After any session that adds new infrastructure: new modules, new data flows, new config.
- After extracting a new shared/canonical helper: confirm nothing else still re-implements the same concern.
- Before a milestone, as a final pass.
Not every session, and not on the working-tree diff. That cadence is the guard skill's job; this one is the
periodic whole-codebase sweep.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

