agentleFS
CRELLIA-S-LCRELLIA-S-L/srs-dd/.claude/skills/srs-arch-audit/SKILL.md

srs-arch-audit

Skill0 starsChanged 33 days agoMITView on GitHub

Audit the architecture of a project that has one — the cut and the forms inside its parts against the specification, against the code, against what it stands on, and against what it costs — and end every finding in an act. Invoke when the user asks whether the architecture is any good, whether a part should be split or merged, or for a review of the cut. For writing or redrawing a part use srs-arch; for recovering a cut that was never written down use srs-arch-harvest.

The file

We split this skill into the rules it gives an agent

2 rules. About 0 are common, 0 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Every element and requirement named to a person is cited at its first mention, as AGENTS.md asks: python3 srs-dd/tools/srs_arch.py --cite …

    line 12
  2. Very specific:

    Every form a part holds, against the same readings and ending in the same acts: the obligations it answers for, the boundary its shape …

    line 28

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md
<!-- Copyright (c) 2026 CRELLIA S.L. — part of SRS-DD, distributed under the MIT License -->

Auditing an architecture

The format lives in srs-dd/arch.md and is not restated here. This audit reports; it changes no record. What to do about a finding is the maintainer's, and a record edited "while we were in there" is a decision nobody took.

Every element and requirement named to a person is cited at its first mention, as AGENTS.md asks: python3 srs-dd/tools/srs_arch.py --cite <ID>…, python3 srs-dd/tools/srs_view.py --cite <ID>….

What to read, in this order

  1. The mechanical state. python3 srs-dd/tools/srs_arch.py --no-write. Do not re-report what it reports: a carrier no part claims, a realized requirement no part carries, a part answering to nothing, an undeclared or forbidden dependency, a circle, a part entered by a name its interface does not state, a part standing on refuted ground. These are the checker's findings and they are already said.
  2. The document. python3 srs-dd/tools/srs_arch_doc.py — its two readings nothing else gives you: where the code disagrees with the model, and what the layer leaves unsaid.
  3. What the cut costs. python3 srs-dd/tools/srs_arch.py --price parts, then --price areas and --price proposed. Links inside a group against links crossing it: a part with few inside and many crossing is a line drawn through the middle of something, and a part whose numbers are far worse than what the areas or the proposal would give is the one to argue about.
  4. What it drives. python3 srs-dd/tools/srs_arch.py --drivers — the requirements that constrain structure. A driver spread across four parts is a finding; so is a part that answers to no driver at all.
  5. What it stands on, where the project keeps a register: the document's section on the ground under each part. A part whose protected obligations rest on something assumed is a part to redraw gently; one on refuted ground is the checker's finding already.

The five readings

Each part, against the readings .claude/skills/srs-arch/SKILL.md carries: the responsibility that needs no "and", the obligations it answers for, the boundary its interface states, what it protects, and what it depends on. A part failing one of them is renamed, split or merged — never documented into looking better.

The form inside a part

Every form a part holds, against the same readings and ending in the same acts: the obligations it answers for, the boundary its shape states, the state it owns alone, and what it is composed of. The checker's comparisons with the code — a symbol nothing declares, a type no form claims, a form entered past its operations, a design the code has caught up with — are already said and are not re-reported. What an audit adds is whether the form is a boundary at all: a record per class is the failure this grain invites, and it passes every check.

How a finding is written

Three parts, and then the act: what the records say, what the code or the numbers say, where they diverge — and then one of rename, split, merge, redraw, or leave with the reason. A finding with no act is an observation, and an observation is worked out or dropped rather than handed over.

No score. Not a number out of ten, not a grade, not a health percentage: two runs would disagree about a cut nobody touched, and a reader would argue with the number instead of the cut. What may stand beside a finding is a count from --price, which anybody can re-derive.

Leaving a part alone is an act. On a system that works it is the commonest one, and an audit that reports only complaints reads as a list of them.

Where it ends

At the report, read once before it is sent and stripped of what the maintainer cannot act on. A redraw the maintainer accepts goes through .claude/skills/srs-arch/SKILL.md — the move record, the ADR, the records themselves — and not from here.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.