agentleFS
shawnclyborshawnclybor/clybor-claude-tooling/.claude/skills/build-validate/SKILL.md

build-validate

Skill6 starsChanged 28 days agoMITView on GitHub

Review a build's PREMISES — the three-to-five things that must be true about the world for the design to be right — once, after build-probe and before build-plan. Not a plan review. Not a per-round gate. Spawns 2-3 reviewers against the premise page and the probe run's UNPROVABLE list, and asks one question: is the frame right. Use when a build's frame is new or contested — a new module whose oracle is not obvious, a schema change that alters what a field MEANS, or a reversal like "we check their sheet" becoming "we produce it". Triggers — "validate the premises", "is the frame right", "check my assumptions before I plan", "/build validate". Do NOT use on a plan, on a repair round, or on a running system.

The file

We split this skill into the rules it gives an agent

2 rules. About 0 are common, 1 is distinctive, and 1 is 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.

  • 1

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Run once, when a build's frame is new or contested:

    line 51
  2. Very specific:

    If you can state every premise in one line and none is in dispute, skip this stage and say so in state.json.

    line 58

≈ “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

Build Validate — the premise check

One question: is the frame right?

Not "is the plan complete", not "is this number correct", not "could a careless implementer game this criterion". Those have cheaper detectors and this stage must not spend itself on them.

Why this is scoped so narrowly — the measurement

Twenty-one recorded rounds across seven slugs under the old design, which reviewed plans:

round-1 must-fix:  8, 13, 14, 15, 12, 12, 9      mean 11.9, every time
total must-fix:    175
self-inflicted:     78  = 45%      (a finding in text the PREVIOUS round's repair wrote)

Three things follow, and the design is built on them.

A freshly-written plan carries ~12 must-fix regardless of care. Seven independent first rounds, range 8–15. That is not a quality signal; it is what happens when prose asserts things about code nobody has run.

Later rounds mostly grade their own damage. Repair rewrites the document, the panel re-reads fresh surface, forever. One slug reached zero must-fix at round 6, took on new scope, and was back to 12 at round 7 — with review-drift.py scoring 12 of 12 self-inflicted.

Sorting one real round by cheapest detector, only 3 of 15 findings needed judgment. Seven were checkable facts (a probe settles those in seconds). Five were plan incoherence (check-plan-soundness.py settles those in under a second). Three were frame errors — and those three were worth the whole exercise, because nothing else finds them.

So: probe the facts, gate the structure mechanically, and spend this stage only on the frame. Premises do not change when you fix a typo, which is precisely why reviewing them cannot spiral.

When to use

Run once, when a build's frame is new or contested:

  • a new module whose oracle is not obvious — what makes this output right?
  • a schema change that alters what an existing field means
  • a reversal of direction — "we audit their sheet" becoming "we generate it"
  • any UNPROVABLE entry that build-probe carried forward

If you can state every premise in one line and none is in dispute, skip this stage and say so in state.json. Skipping a premise check on settled ground is correct, not a shortcut.

When NOT to use — each of these has a cheaper detector

Not thisUse instead
A plan's task listcheck-plan-soundness.py — 8 mechanical classes, under a second
A number, count, filename or line referencebuild-probe
A repair round after findings were fixedNothing. Re-run the probes and the mechanical gate
A running system with a mutation-backed suiteThe suite. Run the thing
A per-iteration code changebuild-execute's conditional review — and it is conditional

Where the rest lives (read the file for the phase you are in)

Phase / topicRead
Phase 1 premise page, Phase 2 pass condition, Phase 3 review agentsreferences/premise-page-and-review.md
Phase 4 record and report, the stop rule, Output, Guidelinesreferences/record-stop-rule-and-output.md

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.