test(helpers): shared skill-census helper with three explicit counts

physicalSkillFiles (symlinked dirs included, root router included),
authoredSkills (realpath-deduped, router excluded), registryEntries
(what ./setup registers: unique frontmatter names + _gstack-command).

One counting authority for the hermetic seeder, context-bill ground
truth, and the catalog-budget test — connect-chrome's dir symlink and
the root router otherwise produce three subtly different hand-rolled
censuses. Ported-wave foundation (C11).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
Garry Tan 2026-08-12 10:50:45 -07:00
parent d078622b73
commit e78db4e309
No known key found for this signature in database
GPG Key ID: C1F69E85C74EFE1D
2 changed files with 165 additions and 0 deletions

View File

@ -0,0 +1,96 @@
/**
* Shared skill census the ONE place that counts skills (C11).
*
* Three consumers (hermetic seeding, context-bill ground truth, catalog-budget
* test) each need a DIFFERENT count of "the skills", and hand-rolled walks
* encode the wrong one somewhere. The counts diverge because of two facts:
*
* 1. `connect-chrome/` is a directory SYMLINK to `open-gstack-browser/`.
* `Dirent.isDirectory()` is false for it (scripts/discover-skills.ts
* skips it) while setup's trailing-slash shell glob follows it.
* 2. The root `SKILL.md` is a router, registered by `setup` under the
* alias name `_gstack-command`, not as an authored skill.
*
* So: use `physicalSkillFiles` when you mean "every SKILL.md a filesystem
* walk can reach" (context-bill's walker), `authoredSkills` when you mean
* "distinct skills a human maintains" (catalog budget), and
* `registryEntries` when you mean "what a host discovers after ./setup"
* (hermetic seeding must mirror this exactly).
*/
import * as fs from 'fs';
import * as path from 'path';
const SKIP = new Set(['node_modules', '.git', 'dist']);
export interface SkillCensus {
/** Every reachable `<dir>/SKILL.md` (symlinked dirs INCLUDED) plus the
* root `SKILL.md` router. Relative paths from root. */
physicalSkillFiles: string[];
/** Distinct authored skills: symlink-deduped by realpath, root router
* EXCLUDED. Each entry is the canonical directory name. */
authoredSkills: string[];
/** What `setup` registers into `~/.claude/skills/` with prefixing off:
* one entry per unique frontmatter `name:` (falling back to dir name),
* plus the `_gstack-command` root alias when the root router exists.
* Symlinked dirs collapse here because they share a frontmatter name. */
registryEntries: string[];
}
/** First `name:` from SKILL.md frontmatter, mirroring `setup`'s
* `grep -m1 '^name:'` (whitespace stripped). Empty string if absent. */
export function frontmatterName(skillMdPath: string): string {
let body: string;
try {
body = fs.readFileSync(skillMdPath, 'utf-8');
} catch {
return '';
}
const m = body.match(/^name:\s*(.+)$/m);
return m ? m[1].replace(/\s+/g, '') : '';
}
export function skillCensus(root: string): SkillCensus {
const physicalSkillFiles: string[] = [];
const authoredByRealpath = new Map<string, string>();
const registrySet = new Set<string>();
if (fs.existsSync(path.join(root, 'SKILL.md'))) {
physicalSkillFiles.push('SKILL.md');
registrySet.add('_gstack-command');
}
for (const entry of fs.readdirSync(root, { withFileTypes: true })) {
if (entry.name.startsWith('.') || SKIP.has(entry.name)) continue;
const dirPath = path.join(root, entry.name);
// Follow directory symlinks like setup's shell glob does; Dirent alone
// reports symlinks as non-directories.
let isDir = entry.isDirectory();
if (!isDir && entry.isSymbolicLink()) {
try {
isDir = fs.statSync(dirPath).isDirectory();
} catch {
continue; // dangling symlink
}
}
if (!isDir) continue;
const skillMd = path.join(dirPath, 'SKILL.md');
if (!fs.existsSync(skillMd)) continue;
physicalSkillFiles.push(`${entry.name}/SKILL.md`);
const real = fs.realpathSync(dirPath);
if (!authoredByRealpath.has(real)) {
authoredByRealpath.set(real, path.basename(real));
}
registrySet.add(frontmatterName(skillMd) || entry.name);
}
return {
physicalSkillFiles: physicalSkillFiles.sort(),
authoredSkills: [...authoredByRealpath.values()].sort(),
registryEntries: [...registrySet].sort(),
};
}

69
test/skill-census.test.ts Normal file
View File

@ -0,0 +1,69 @@
/**
* Pins the three-count contract of test/helpers/skill-census.ts (C11).
*
* No hardcoded totals here the catalog-budget test owns the ratchet.
* This file pins the STRUCTURAL relationships that make the three counts
* mean different things, using the live repo as the fixture.
*/
import { describe, expect, it } from 'bun:test';
import * as fs from 'fs';
import * as path from 'path';
import { frontmatterName, skillCensus } from './helpers/skill-census';
const ROOT = path.resolve(new URL(import.meta.url).pathname, '..', '..');
const census = skillCensus(ROOT);
describe('skillCensus', () => {
it('physicalSkillFiles includes the root router and the symlinked dir', () => {
expect(census.physicalSkillFiles).toContain('SKILL.md');
expect(census.physicalSkillFiles).toContain('connect-chrome/SKILL.md');
expect(census.physicalSkillFiles).toContain('open-gstack-browser/SKILL.md');
});
it('authoredSkills dedupes the connect-chrome symlink and excludes the root router', () => {
expect(census.authoredSkills).toContain('open-gstack-browser');
expect(census.authoredSkills).not.toContain('connect-chrome');
// Root router is not an authored skill; its dir entry would be '' anyway.
for (const name of census.authoredSkills) expect(name.length).toBeGreaterThan(0);
});
it('registryEntries carries the root alias and collapses shared frontmatter names', () => {
expect(census.registryEntries).toContain('_gstack-command');
expect(
census.registryEntries.filter((n) => n === 'open-gstack-browser'),
).toHaveLength(1);
});
it('count relationships hold: physical = authored + root + symlink dups', () => {
const symlinkDups = census.physicalSkillFiles.length - 1 - census.authoredSkills.length;
expect(symlinkDups).toBeGreaterThanOrEqual(1); // connect-chrome today
// Registry = unique frontmatter names + root alias. It can only collapse
// entries relative to physical, never invent them.
expect(census.registryEntries.length).toBeLessThanOrEqual(census.physicalSkillFiles.length);
expect(census.registryEntries.length).toBeGreaterThan(census.authoredSkills.length - 1);
});
it('frontmatterName mirrors setup: first ^name: line, whitespace stripped', () => {
const qa = frontmatterName(path.join(ROOT, 'qa', 'SKILL.md'));
expect(qa).toBe('qa');
const alias = frontmatterName(path.join(ROOT, 'connect-chrome', 'SKILL.md'));
expect(alias).toBe('open-gstack-browser');
expect(frontmatterName(path.join(ROOT, 'no-such-dir', 'SKILL.md'))).toBe('');
});
it('every registry entry a host would see resolves back to a physical SKILL.md', () => {
const names = new Set(
census.physicalSkillFiles
.filter((p) => p !== 'SKILL.md')
.map((p) => frontmatterName(path.join(ROOT, p)) || path.dirname(p)),
);
for (const entry of census.registryEntries) {
if (entry === '_gstack-command') {
expect(fs.existsSync(path.join(ROOT, 'SKILL.md'))).toBe(true);
} else {
expect(names.has(entry)).toBe(true);
}
}
});
});