agentleFS
tatsoft-llctatsoft-llc/FrameworX-Projects/.claude/skills/review-the-ai-build/SKILL.md

review-the-ai-build

Skill1 starsChanged 28 days agoApache-2.0View on GitHub

Independently review a FrameworX solution that the AI just built, against its spec, before you trust it. Use after a build or a build layer completes, when the user says "review the build", "check what you built", "did this match the spec", "audit the solution", or before running the tests on a fresh build. Reads the real solution over the Designer or Console MCP and the running runtime, compares to Sites/<Name>/Specifications/, and reports findings by severity. It reviews; it does not fix. Distinct from run-tests (which mechanically asserts expected values): this is the human-judgment pass that catches structure, naming, anti-patterns, and reachability problems a fixed assertion list would miss.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 1 is distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    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:

    Read Sites/<Name>/Specifications/ end to end.

    line 39
  2. Distinctive:

    Before line-level detail, ask whether the build is the right shape:

    line 44
  3. Very specific:

    If the review was blocked or degraded by a tool gap (a value you could not read, a state the MCP does not expose), record it in an ## MCP …

    line 102

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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

Review the AI build

The AI built a solution. Before anyone trusts it, look at what was actually built, against what the spec asked for, with fresh eyes. This is the judgment companion to the mechanical test harness. It finds the things a fixed assertion list does not: a foundation that drifted from the spec, an anti-pattern, a screen that binds to nothing, an option that was written but is not reachable.

This skill reviews. It does not fix. It produces a findings list. The user decides what to do with it.

Before you start

  • Confirm a spec exists at Sites/<Name>/Specifications/. The review compares the build to the spec. With no spec, say so and review against stated intent instead, flagging that there is no contract.
  • Bootstrap the MCP first (open_solution / open_workspace). Re-bootstrap if you switched solutions.
  • Review the ai-test- sandbox, not a real solution.

The review, in four passes

Pass 0: ground

Read Sites/<Name>/Specifications/ end to end. Note the acceptance checks in Requirements.md; they are the spine of the review. Read KnowledgeBase/References/frameworx-vocabulary.md if any term is unclear.

Pass 1: scope and boundary

Before line-level detail, ask whether the build is the right shape:

  • Does the foundation (UserTypes, UNS, asset tree) match the spec's, in count and in name? A foundation that drifted is a containment problem, not a detail to patch.
  • Is anything built that the spec did not ask for? Scope creep in a solution is as real as in code.
  • Is each layer present: foundation, connect, monitor, screens? A missing layer is the headline.

A build that is line-clean but structurally wrong is not approved. Scope first.

Pass 2: correctness and anti-patterns

Now the detail. Use get_objects and browse_object_model to read the real objects.

  • Foundation: UDT members, units, ranges, retentive flags match the spec. Names are PascalCase and hierarchical.
  • Connect: tags have a source (a device point or a simulation). A tag with no source holds no value.
  • Alarms: the right tags are alarmed, grouped, and area-assigned. Watch for alarm fatigue (everything alarmed, one threshold repeated everywhere).
  • Historian: the right tags are logged, not all of them. Watch for historize-everything.
  • Displays: controls are real (list_elements), bindings resolve, screens are built from Symbols rather than copy-paste. Take a get_screenshot of at least the main display; the render is ground truth, and an empty binding shows as a default placeholder, not an error.
  • Orphans: references to objects that no longer exist. Spot-check bindings with get_objects and confirm each referenced tag, display, or symbol actually resolves.

Pass 3: reachability in the runtime

A written row is not a working feature. Start the runtime (Development profile) if it is not up, and confirm with the Runtime tools:

  • Tags named in the spec hold live values with good quality and recent timestamps.
  • An alarm goes active when its condition is forced true.
  • History accumulates for a logged tag.

Report

Findings by severity. For each: the exact object (table, name, member), what the spec said, what the build has, and why it matters. Do not paraphrase; cite the object.

  • MAJOR: structure or behavior that contradicts the spec or breaks at runtime. Must be fixed before the build is trusted.
  • MINOR: drift, naming, or an anti-pattern that is not yet breaking but will cost later.
  • NOTE: an observation or a suggestion, no action required.

Findings persist to the project record, not just the transcript: write the list as a dated table in Sites/<Name>/Documentation/ (for example Review-Findings-<date>.md), one row per finding with finding / severity / object / status. Later sessions work from that table, not from a transcript, and write their verdicts back into it, updating each row's status in place: fixed, refuted-on-re-check, or open. Rows are never deleted; the table is the review-side twin of the commissioning record, evidence with a status trail rather than a one-time report.

End with one line: approved, or not approved with the count of MAJOR findings. If you reviewed against intent because there was no spec, say so. Then stop. Fixing is a separate, user-directed step.

MCP findings

If the review was blocked or degraded by a tool gap (a value you could not read, a state the MCP does not expose), record it in an ## MCP findings block per KnowledgeBase/Method/Solution-Build-Method.md §8.

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.