scripts/

The guards. Every claim this standard makes about your repository is checked by something in here, and all of it runs on Node built-ins alone - no install step, no dependency to audit, nothing to break when your lockfile changes.

That constraint is deliberate. A compliance tool that needs its own dependency tree is a compliance tool that eventually cannot run, and it will be the moment you most need the answer.

What is in here, and what each one refuses to let you do

scripts/spec/ holds the spec engine the skills invoke - the gate, the setup and the templates.

What it is for

So that "we follow a standard" becomes a number. self-verify reports drift: how many required entries are unmet. Zero means compliant against the manifest, and any number above it comes with the list. That is a claim a reviewer can check in eight seconds without trusting anybody.

How you actually use it

node scripts/self-verify.mjs

It exits non-zero on any failure, which is what makes it usable as a CI gate rather than as a report somebody reads.

What does not go in here

Your product's tooling. These are the standard's guards; your build scripts live wherever your stack puts them.

Anything with a dependency. Node built-ins only. This is not minimalism for its own sake - see above.

A check that cannot explain its failure. Every guard prints what failed and where. A red build that does not say why gets bypassed, then permanently disabled.

What drift 0 does not certify

Worth being blunt, because the number is the product's headline claim. self-verify checks the mechanical tier: does the file exist, does the heading exist, does the guard pass. It cannot check whether your decision records actually record your decisions, or whether a spec that names a persona genuinely serves them. That is the judgment tier and it stays at review.

Drift 0 with empty shells is a hollow win. The placeholder scan warns about it and deliberately does not count it as drift, because turning judgment into an integer is how a metric starts being gamed.

Decisions behind it

• ADR-005 - the manifest is the contract and the engine only reads it. Hard-coding the checks was the alternative, and it means every new entry in the standard is an engine change, a release and an upgrade. Manifest content is data; only a change to how an entry is interpreted is a change to the tool.
• Dependency-free, permanently. A single well-chosen library would make several of these shorter. It would also make them installable, versioned and breakable.
• Warnings never count as drift. Making the placeholder scan drift-bearing was tried in spirit and rejected: it converts "we have not filled this in yet" into a build failure, and the fix people reach for is deleting the file.

What is inside

• generate-dashboard/ - renders the work state - pool, sprints, timeline, records, changelog - as one static page for the people who never open the repository; index.mjs is the generator, src/ the page and password-gate material it inlines. A projection, never a second place work is tracked
• lib/ - what more than one guard needs - glob matching, so two guards cannot answer the same question differently
• spec/ - the spec engine runtime the skills invoke (gate, setup, templates, MIT licence)
• adoption-gates.mjs - the Gate 2 and Gate 5 artifacts carry what the gate is for - eight passes rated, the scope block's total equal to the sum of its categories, an owner role on every alignment item
• backlog-archive-check.mjs - a closed row reached the archive with a pointer to what its content became, instead of being deleted (ADR-051)
• decision-records-check.mjs - the ADR/BDR README index and the files on disk agree - no duplicate id, nothing indexed that is not there, nothing there that is not indexed
• facts-check.mjs - a fact restated in prose still agrees with its source - declared homes and claims (R4)
• schema-pair.mjs - the DDL under database/schema/ and its typed twin stay 1:1 - the declared edge resolves and every name is covered
• self-verify.mjs - the verify step - checks the repo against this manifest
• spec-guard.mjs - code<->spec coupling guard (+ --audit) - the tool ships at core, the blocking CI gate is scale (R11)
• spec-structure.mjs - spec layout guard
• sprint-guard.mjs - proves the one-place invariant the pool and the sprints depend on
• verifyAgentGuards.sh - drives every guard with real commands - they only emit output on a denial, so a broken guard is otherwise silent

What checks it

• node scripts/self-verify.mjs - compliance with the standard Blocks the build when it fails.
• node scripts/spec-structure.mjs --block - spec files are where and shaped as the standard requires Blocks the build when it fails.
• node scripts/facts-check.mjs - declared facts still agree with their source (skips itself where none are declared) Blocks the build when it fails.
• node scripts/schema-pair.mjs --block - the schema and its typed twin have not drifted (skips itself where there is no database) Blocks the build when it fails.
• node scripts/spec-guard.mjs --block - domain code changed without touching its spec fails CI Blocks the build when it fails.
• node scripts/sprint-guard.mjs --block - an intent is in the backlog pool or in exactly one sprint (skips itself where there are no sprints) Blocks the build when it fails.
• node scripts/decision-records-check.mjs --block - the ADR/BDR index and the directory on disk agree - duplicate id, orphaned index row, or unindexed record all fail Blocks the build when it fails.
• node scripts/adoption-gates.mjs --block - the two gate artifacts carry what the gate is for - eight passes rated, the scope block adds up, every alignment item owned (skips itself where no assessment exists, which self-verify already reports) Blocks the build when it fails.
• node scripts/backlog-archive-check.mjs --block - a row removed from the pool reached the archive, and every archived row names where its content went (skips its archive checks where nothing has been archived yet) Blocks the build when it fails.

Read this page's markdown, or its source on GitHub.