@event4u/agent-config 14.8.0 → 14.10.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/.claude-plugin/marketplace.json +1 -1
- package/CHANGELOG.md +214 -237
- package/README.md +1 -1
- package/agents/templates/.ai-council.yml.example +34 -4
- package/dist/agent-src/commands/pr/merge.md +9 -5
- package/dist/agent-src/commands/review/changes.md +95 -14
- package/dist/agent-src/commands/roadmap/process-full.md +2 -2
- package/dist/agent-src/contexts/execution/host-capability-manifest.md +77 -0
- package/dist/agent-src/contexts/execution/roadmap-execution-contract.md +36 -0
- package/dist/agent-src/contexts/execution/subagent-response-contract.md +80 -0
- package/dist/agent-src/rules/untrusted-input-defense.md +9 -3
- package/dist/agent-src/skills/async-python-patterns/SKILL.md +1 -1
- package/dist/agent-src/skills/code-refactoring/SKILL.md +1 -1
- package/dist/agent-src/skills/code-review/evals/evals.json +87 -46
- package/dist/agent-src/skills/condense-memory/SKILL.md +1 -1
- package/dist/agent-src/skills/conventional-commits-writing/SKILL.md +1 -1
- package/dist/agent-src/skills/design-review/SKILL.md +1 -1
- package/dist/agent-src/skills/discovery-interview/SKILL.md +1 -1
- package/dist/agent-src/skills/doc-coauthoring/SKILL.md +1 -1
- package/dist/agent-src/skills/docx-authoring/SKILL.md +1 -1
- package/dist/agent-src/skills/editorial-calendar/SKILL.md +1 -1
- package/dist/agent-src/skills/github-ci/SKILL.md +1 -1
- package/dist/agent-src/skills/humanizer/SKILL.md +39 -0
- package/dist/agent-src/skills/humanizer/evals/strip_fixtures.json +57 -0
- package/dist/agent-src/skills/humanizer/references/fixtures.md +54 -0
- package/dist/agent-src/skills/judge-artifact-completeness/SKILL.md +1 -1
- package/dist/agent-src/skills/judge-injection-defense/SKILL.md +1 -1
- package/dist/agent-src/skills/judge-spec-compliance/SKILL.md +154 -0
- package/dist/agent-src/skills/judge-synthesis/SKILL.md +45 -0
- package/dist/agent-src/skills/laravel-validation/SKILL.md +1 -1
- package/dist/agent-src/skills/legal-intake-triage/SKILL.md +1 -1
- package/dist/agent-src/skills/legal-practice-profile/SKILL.md +1 -1
- package/dist/agent-src/skills/messaging-architecture/SKILL.md +1 -1
- package/dist/agent-src/skills/migration-architect/SKILL.md +1 -1
- package/dist/agent-src/skills/module-detect-on-the-fly/SKILL.md +1 -1
- package/dist/agent-src/skills/module-management/SKILL.md +1 -1
- package/dist/agent-src/skills/motion-choreographer/SKILL.md +1 -1
- package/dist/agent-src/skills/onboarding-program/SKILL.md +1 -1
- package/dist/agent-src/skills/perf-feedback-craft/SKILL.md +1 -1
- package/dist/agent-src/skills/prompt-optimizer/SKILL.md +1 -1
- package/dist/agent-src/skills/receiving-code-review/SKILL.md +1 -1
- package/dist/agent-src/skills/roadmap-writing/SKILL.md +18 -18
- package/dist/agent-src/skills/scene-expander/SKILL.md +1 -1
- package/dist/agent-src/skills/secrets-management/SKILL.md +1 -1
- package/dist/agent-src/skills/using-git-worktrees/SKILL.md +1 -1
- package/dist/agent-src/templates/agents/agent-project-settings.example.yml +1 -1
- package/dist/agent-src/templates/roadmaps.md +8 -0
- package/dist/cli/registry.js +2 -0
- package/dist/cli/registry.js.map +1 -1
- package/dist/cli-delegate/{chunk-MMSGVZKV.js → chunk-77GR3VD4.js} +65 -40
- package/dist/cli-delegate/{chunk-Q26HT4QT.js → chunk-E46KPN6K.js} +251 -150
- package/dist/cli-delegate/{chunk-EEJUMCDZ.js → chunk-PLSG6UL5.js} +1 -1
- package/dist/cli-delegate/cmd_conformance.js +89 -13
- package/dist/cli-delegate/cmd_doctor.js +3 -3
- package/dist/cli-delegate/cmd_handoff.js +2 -2
- package/dist/cli-delegate/cmd_memory_get.js +1 -1
- package/dist/cli-delegate/cmd_route_audit.js +1 -1
- package/dist/cli-delegate/cmd_settings_get.js +1 -1
- package/dist/cli-delegate/cmd_settings_set.js +1 -1
- package/dist/discovery/deprecation-report.md +1 -1
- package/dist/discovery/discovery-manifest.json +64 -42
- package/dist/discovery/discovery-manifest.json.sha256 +1 -1
- package/dist/discovery/discovery-manifest.summary.md +3 -3
- package/dist/discovery/dormancy-report.md +1 -1
- package/dist/discovery/orphan-report.md +1 -1
- package/dist/discovery/packs.json +6 -5
- package/dist/discovery/trust-report.md +2 -2
- package/dist/discovery/workspaces.json +5 -4
- package/dist/hooks/dispatch.js +562 -204
- package/dist/mcp/registry-manifest.json +3 -3
- package/dist/mcp/server.json +2 -2
- package/dist/mcp/server.mjs +808 -331
- package/dist/scripts/_lib/claude_plugin.js +162 -0
- package/dist/scripts/_lib/claude_plugin.js.map +1 -0
- package/dist/scripts/_lib/environment_detector.js +397 -0
- package/dist/scripts/_lib/environment_detector.js.map +1 -0
- package/dist/scripts/_lib/spawn_env.js +129 -0
- package/dist/scripts/_lib/spawn_env.js.map +1 -0
- package/dist/scripts/ai_council/modes.js +199 -0
- package/dist/scripts/ai_council/modes.js.map +1 -0
- package/dist/scripts/ai_council/transport_resolver.js +418 -0
- package/dist/scripts/ai_council/transport_resolver.js.map +1 -0
- package/dist/server/routes/wizard.js +3 -2
- package/dist/server/routes/wizard.js.map +1 -1
- package/dist/ui/assets/index-Cxgy63kz.js +26 -0
- package/dist/ui/assets/index-Cxgy63kz.js.map +1 -0
- package/dist/ui/index.html +1 -1
- package/docs/guidelines/agent-infra/traceability-field-mechanics.md +50 -0
- package/docs/guidelines/agent-infra/untrusted-input-spotlighting.md +17 -0
- package/package.json +7 -3
- package/src/config/estate-count-budget.json +13 -744
- package/src/config/evaluator-budgets.json +4 -4
- package/src/config/gate-coverage.yml +267 -9
- package/src/config/gate-violation-baselines.json +72 -72
- package/src/config/preamble-payload-budget.json +71 -67
- package/src/config/publish-surface.json +60 -0
- package/src/config/requirements-trace-budget.json +32 -0
- package/src/config/rule-stub-ceilings.json +8 -1
- package/src/scripts/_dispatch.bash +7 -0
- package/src/scripts/_lib/base_tree.ts +175 -0
- package/src/scripts/_lib/billing_grant.ts +165 -0
- package/src/scripts/_lib/billing_grant_cli.ts +123 -0
- package/src/scripts/_lib/council_fallback_posture.ts +53 -6
- package/src/scripts/_lib/council_fallback_wiring.ts +61 -1
- package/src/scripts/_lib/host_capability.ts +59 -5
- package/src/scripts/_lib/quota_parked.ts +133 -0
- package/src/scripts/_lib/retrieval_sanitize.ts +62 -0
- package/src/scripts/_lib/review_telemetry.ts +196 -0
- package/src/scripts/_lib/subagent_response.ts +92 -0
- package/src/scripts/adr/regenerate_index.ts +14 -7
- package/src/scripts/ai_council/clients.ts +12 -1
- package/src/scripts/ai_council/config.ts +72 -115
- package/src/scripts/ai_council/fallback_config.ts +19 -7
- package/src/scripts/ai_council/mid_flight_fallback.ts +88 -7
- package/src/scripts/ai_council/orchestrator.ts +2 -2
- package/src/scripts/ai_council/py_format.ts +124 -0
- package/src/scripts/ai_council/seating.ts +211 -0
- package/src/scripts/ai_council/transport_resolver.ts +65 -2
- package/src/scripts/audit_adr_coverage.ts +33 -6
- package/src/scripts/check_council_pin_staleness.ts +300 -0
- package/src/scripts/check_estate_count.ts +420 -293
- package/src/scripts/check_gate_coverage.ts +91 -0
- package/src/scripts/check_no_stub_inventory_table.ts +292 -0
- package/src/scripts/check_pack_passport_reconciliation.ts +274 -0
- package/src/scripts/check_pack_size.ts +234 -1
- package/src/scripts/check_publish_surface.ts +297 -0
- package/src/scripts/check_requirements_trace.ts +378 -0
- package/src/scripts/check_secret_leak.ts +166 -12
- package/src/scripts/code_graph/cli.ts +44 -5
- package/src/scripts/code_graph/extract.ts +87 -0
- package/src/scripts/council_cli.ts +12 -16
- package/src/scripts/detect_ai_tells.ts +87 -0
- package/src/scripts/dispatch_r2_reviewer.ts +42 -0
- package/src/scripts/generate_capabilities_index.ts +41 -0
- package/src/scripts/generate_pack_manifests.ts +78 -0
- package/src/scripts/git_authorization_hook.ts +54 -4
- package/src/scripts/hooks/block_unauthorized_git.ts +19 -1
- package/src/scripts/hooks/run_continuation_hook.ts +3 -1
- package/src/scripts/hooks/subagent_ledger_hook.ts +76 -11
- package/src/scripts/injection_scan_hook.ts +120 -10
- package/src/scripts/lint_adr_sweep_routing.ts +365 -0
- package/src/scripts/lint_workflow_security.ts +53 -10
- package/src/scripts/lint_workflow_security_allowlist.json +2 -18
- package/src/scripts/report_envelope_rate.ts +166 -0
- package/src/scripts/review_axis_report.ts +218 -0
- package/src/scripts/run_supervise.ts +46 -1
- package/src/scripts/schemas/evals.schema.json +5 -0
- package/src/scripts/schemas/pack.schema.json +62 -4
- package/src/scripts/schemas/skill.schema.json +2 -2
- package/src/scripts/sync_pr_branch.ts +14 -7
- package/src/scripts/tighten_skill_descriptions.ts +264 -0
- package/dist/cli/commands/commands.candidates.test.js +0 -257
- package/dist/cli/commands/commands.candidates.test.js.map +0 -1
- package/dist/cli/commands/profiles.test.js +0 -88
- package/dist/cli/commands/profiles.test.js.map +0 -1
- package/dist/cli/commands/recordTriggerEval.test.js +0 -113
- package/dist/cli/commands/recordTriggerEval.test.js.map +0 -1
- package/dist/cli/configRoot.test.js +0 -97
- package/dist/cli/configRoot.test.js.map +0 -1
- package/dist/cli/mcp/content.test.js +0 -54
- package/dist/cli/mcp/content.test.js.map +0 -1
- package/dist/cli/mcp/dispatch.test.js +0 -108
- package/dist/cli/mcp/dispatch.test.js.map +0 -1
- package/dist/server/writeRoot.test.js +0 -91
- package/dist/server/writeRoot.test.js.map +0 -1
- package/dist/shared/capabilities.test.js +0 -39
- package/dist/shared/capabilities.test.js.map +0 -1
- package/dist/ui/assets/index-C1pBRm1f.js +0 -26
- package/dist/ui/assets/index-C1pBRm1f.js.map +0 -1
- /package/dist/cli-delegate/{chunk-N7CDDGI3.js → chunk-CYH5SP6M.js} +0 -0
|
@@ -0,0 +1,274 @@
|
|
|
1
|
+
#!/usr/bin/env node
|
|
2
|
+
/**
|
|
3
|
+
* Do the per-pack passports add up to the census the tree already runs?
|
|
4
|
+
*
|
|
5
|
+
* A per-pack number published beside a tree-wide number it contradicts is worse
|
|
6
|
+
* than no per-pack number: both get quoted, and nobody can say which is wrong.
|
|
7
|
+
* This is the check that says so out loud.
|
|
8
|
+
*
|
|
9
|
+
* THE ±10 % BAND THIS WAS SPECIFIED WITH RESTED ON A FALSE PREMISE, and the
|
|
10
|
+
* measurement is what refuted it. The band assumed packs roughly PARTITION the
|
|
11
|
+
* tree, so per-pack sums should approach the tree-wide census. Measured on
|
|
12
|
+
* 2026-08-22: across all 17 packs, `dependencies` claims **4 skills and 0
|
|
13
|
+
* rules**, against **291 skills and 118 rules** in the projection. Packs claim
|
|
14
|
+
* ~1 % of the artefacts the census counts.
|
|
15
|
+
*
|
|
16
|
+
* So the two numbers are not a reconciliation pair at all:
|
|
17
|
+
*
|
|
18
|
+
* census — the PROJECTED tree (`dist/agent-src/`), everything that ships
|
|
19
|
+
* passport — what a pack CLAIMS, which is a small, deliberate subset
|
|
20
|
+
*
|
|
21
|
+
* A gate failing on an 84 % gap would be failing on a property of the tree, not
|
|
22
|
+
* on a defect — and a gate that is red by construction gets disabled rather
|
|
23
|
+
* than fixed. So this REPORTS coverage as a first-class number and exits 0.
|
|
24
|
+
* That is exactly what the step behind it asked for: a passport that does not
|
|
25
|
+
* reconcile "must say what, in-band, rather than being published alongside a
|
|
26
|
+
* number it contradicts."
|
|
27
|
+
*
|
|
28
|
+
* What would make it a gate: pack coverage approaching the tree. Then the band
|
|
29
|
+
* becomes meaningful and the exit code can follow it. Until then the useful
|
|
30
|
+
* output is the RANKING plus the coverage figure.
|
|
31
|
+
*
|
|
32
|
+
* Both sides are `chars/4` — the basis `preamble-payload-budget.json` declares
|
|
33
|
+
* for itself. Mixing an exact tokenizer into one side would make the gap
|
|
34
|
+
* between two METHODS read as a discrepancy in the packs.
|
|
35
|
+
*
|
|
36
|
+
* Also emits the RANKING, because the useful output is not the numbers but
|
|
37
|
+
* which packs carry the weight: a per-pack table sorted by cost turns "the
|
|
38
|
+
* estate grew" into "this pack grew".
|
|
39
|
+
*
|
|
40
|
+
* Exit: 0 always (it is a REPORT, not a gate), 2 on a dead scan scope. See the
|
|
41
|
+
* band note above for why the exit code does not follow the gap.
|
|
42
|
+
*/
|
|
43
|
+
import * as fs from 'node:fs';
|
|
44
|
+
import * as path from 'node:path';
|
|
45
|
+
import { fileURLToPath } from 'node:url';
|
|
46
|
+
|
|
47
|
+
import { assertScanned, DeadScopeError } from './_lib/scan_scope.js';
|
|
48
|
+
import { runSelfTest } from './_lib/gate_self_test.js';
|
|
49
|
+
import { censusRuleDir, censusSkillsCatalog } from './preamble_byte_census.js';
|
|
50
|
+
|
|
51
|
+
const _HERE = fileURLToPath(import.meta.url);
|
|
52
|
+
export const REPO_ROOT = path.resolve(path.dirname(_HERE), '..', '..');
|
|
53
|
+
export const PACKS_DIR = path.join(REPO_ROOT, 'src/packs');
|
|
54
|
+
export const RANKING = path.join(REPO_ROOT, 'agents/evidence/reports/pack-token-ranking.md');
|
|
55
|
+
|
|
56
|
+
/** ±10 %. See the module docstring for why a band exists at all. */
|
|
57
|
+
export const TOLERANCE = 0.1;
|
|
58
|
+
|
|
59
|
+
export interface Passport {
|
|
60
|
+
pack: string;
|
|
61
|
+
rules_tokens: number;
|
|
62
|
+
catalog_tokens: number;
|
|
63
|
+
commands_tokens: number;
|
|
64
|
+
other_tokens: number;
|
|
65
|
+
total_tokens: number;
|
|
66
|
+
}
|
|
67
|
+
|
|
68
|
+
/** Read `token_passport:` out of a generated pack.yaml — no YAML dependency. */
|
|
69
|
+
export function readPassport(pack: string, text: string): Passport | null {
|
|
70
|
+
const start = text.indexOf('token_passport:');
|
|
71
|
+
if (start < 0) return null;
|
|
72
|
+
const num = (key: string): number => {
|
|
73
|
+
const m = new RegExp(`^\\s+${key}:\\s*(\\d+)\\s*$`, 'm').exec(text.slice(start));
|
|
74
|
+
return m === null ? 0 : Number(m[1]);
|
|
75
|
+
};
|
|
76
|
+
return {
|
|
77
|
+
pack,
|
|
78
|
+
rules_tokens: num('rules_tokens'),
|
|
79
|
+
catalog_tokens: num('catalog_tokens'),
|
|
80
|
+
commands_tokens: num('commands_tokens'),
|
|
81
|
+
other_tokens: num('other_tokens'),
|
|
82
|
+
total_tokens: num('total_tokens'),
|
|
83
|
+
};
|
|
84
|
+
}
|
|
85
|
+
|
|
86
|
+
export function collectPassports(dir: string = PACKS_DIR): Passport[] {
|
|
87
|
+
if (!fs.existsSync(dir)) return [];
|
|
88
|
+
const out: Passport[] = [];
|
|
89
|
+
for (const e of fs.readdirSync(dir, { withFileTypes: true })) {
|
|
90
|
+
if (!e.isDirectory()) continue;
|
|
91
|
+
const f = path.join(dir, e.name, 'pack.yaml');
|
|
92
|
+
if (!fs.existsSync(f)) continue;
|
|
93
|
+
const p = readPassport(e.name, fs.readFileSync(f, 'utf-8'));
|
|
94
|
+
if (p !== null) out.push(p);
|
|
95
|
+
}
|
|
96
|
+
return out;
|
|
97
|
+
}
|
|
98
|
+
|
|
99
|
+
export interface Reconciliation {
|
|
100
|
+
bucket: string;
|
|
101
|
+
packSum: number;
|
|
102
|
+
censusTotal: number;
|
|
103
|
+
gapPct: number;
|
|
104
|
+
inBand: boolean;
|
|
105
|
+
}
|
|
106
|
+
|
|
107
|
+
export function reconcile(packSum: number, censusTotal: number, bucket: string): Reconciliation {
|
|
108
|
+
// Denominator is the CENSUS, which is the side with the independent
|
|
109
|
+
// measurement. Dividing by the pack sum would let a passport that measured
|
|
110
|
+
// almost nothing report a small gap.
|
|
111
|
+
const gapPct = censusTotal === 0 ? (packSum === 0 ? 0 : 1) : Math.abs(packSum - censusTotal) / censusTotal;
|
|
112
|
+
return { bucket, packSum, censusTotal, gapPct, inBand: gapPct <= TOLERANCE };
|
|
113
|
+
}
|
|
114
|
+
|
|
115
|
+
export function renderRanking(ps: readonly Passport[]): string {
|
|
116
|
+
const sorted = [...ps].sort((a, b) => b.total_tokens - a.total_tokens);
|
|
117
|
+
const total = sorted.reduce((a, p) => a + p.total_tokens, 0);
|
|
118
|
+
const lines = [
|
|
119
|
+
'<!-- evidence-type: analysis -->',
|
|
120
|
+
'# Pack token ranking — which packs carry the weight',
|
|
121
|
+
'',
|
|
122
|
+
'GENERATED by `./scripts-run src/scripts/check_pack_passport_reconciliation --write-ranking`.',
|
|
123
|
+
'Do not hand-edit. Basis `chars/4`, per `docs/contracts/pack-token-passport.md`.',
|
|
124
|
+
'',
|
|
125
|
+
'The numbers are each pack\'s STANDING contribution — what is paid whether or',
|
|
126
|
+
'not the pack is used. A skill body loaded on invocation is a runtime cost and',
|
|
127
|
+
'is not here, and the residual bucket (tool definitions + dispatch prompt) is',
|
|
128
|
+
'not attributable to a pack at all.',
|
|
129
|
+
'',
|
|
130
|
+
`**${String(sorted.length)} packs, ${String(total)} tokens total.**`,
|
|
131
|
+
'',
|
|
132
|
+
'| rank | pack | total | rules | catalog | commands | other | share |',
|
|
133
|
+
'|---:|---|---:|---:|---:|---:|---:|---:|',
|
|
134
|
+
];
|
|
135
|
+
sorted.forEach((p, i) => {
|
|
136
|
+
const share = total === 0 ? 0 : (p.total_tokens / total) * 100;
|
|
137
|
+
lines.push(
|
|
138
|
+
`| ${String(i + 1)} | \`${p.pack}\` | ${String(p.total_tokens)} | ${String(p.rules_tokens)} | ` +
|
|
139
|
+
`${String(p.catalog_tokens)} | ${String(p.commands_tokens)} | ${String(p.other_tokens)} | ${share.toFixed(1)} % |`,
|
|
140
|
+
);
|
|
141
|
+
});
|
|
142
|
+
return `${lines.join('\n')}\n`;
|
|
143
|
+
}
|
|
144
|
+
|
|
145
|
+
/**
|
|
146
|
+
* Discrimination over the two pure units — the parser and the ratio.
|
|
147
|
+
*
|
|
148
|
+
* This script exits 0 on every path by design (it is a report, not a gate), so
|
|
149
|
+
* no planted file can make it red and the canary harness cannot prove anything
|
|
150
|
+
* about it. The `scanned:` floor is the one thing about it that CAN break, and
|
|
151
|
+
* a self-test is what proves the two functions behind the number actually
|
|
152
|
+
* discriminate rather than returning something shaped like an answer.
|
|
153
|
+
*/
|
|
154
|
+
function selfTest(): number {
|
|
155
|
+
const yaml = [
|
|
156
|
+
'artefact_count: 3',
|
|
157
|
+
'token_passport:',
|
|
158
|
+
' basis: chars/4',
|
|
159
|
+
' catalog_tokens: 20',
|
|
160
|
+
' commands_tokens: 5',
|
|
161
|
+
' other_tokens: 1',
|
|
162
|
+
' rules_tokens: 10',
|
|
163
|
+
' total_tokens: 36',
|
|
164
|
+
].join('\n');
|
|
165
|
+
return runSelfTest({
|
|
166
|
+
gate: 'check_pack_passport_reconciliation',
|
|
167
|
+
minCases: 6,
|
|
168
|
+
minRejectCases: 3,
|
|
169
|
+
cases: [
|
|
170
|
+
{
|
|
171
|
+
name: 'a manifest with a passport parses every field',
|
|
172
|
+
expect: 'reject',
|
|
173
|
+
run: () => {
|
|
174
|
+
const p = readPassport('x', yaml);
|
|
175
|
+
return p !== null && p.rules_tokens === 10 && p.total_tokens === 36 ? 1 : 0;
|
|
176
|
+
},
|
|
177
|
+
},
|
|
178
|
+
{
|
|
179
|
+
name: 'a manifest with NO passport returns null, not a zero passport',
|
|
180
|
+
expect: 'accept',
|
|
181
|
+
run: () => (readPassport('x', 'artefact_count: 3\n') === null ? 0 : 1),
|
|
182
|
+
},
|
|
183
|
+
{
|
|
184
|
+
name: 'an exact match is in band',
|
|
185
|
+
expect: 'accept',
|
|
186
|
+
run: () => (reconcile(100, 100, 'b').inBand ? 0 : 1),
|
|
187
|
+
},
|
|
188
|
+
{
|
|
189
|
+
name: 'a 9 % gap is in band and an 11 % gap is not',
|
|
190
|
+
expect: 'reject',
|
|
191
|
+
run: () => (reconcile(109, 100, 'b').inBand && !reconcile(111, 100, 'b').inBand ? 1 : 0),
|
|
192
|
+
},
|
|
193
|
+
{
|
|
194
|
+
// The denominator is the CENSUS, the side with the independent
|
|
195
|
+
// measurement. Dividing by the pack sum would let a passport
|
|
196
|
+
// that measured almost nothing report a small gap — which is
|
|
197
|
+
// exactly the failure this repo's own measurement produced.
|
|
198
|
+
name: 'the denominator is the census, not the pack sum',
|
|
199
|
+
expect: 'reject',
|
|
200
|
+
run: () => (Math.abs(reconcile(1, 100, 'b').gapPct - 0.99) < 1e-9 ? 1 : 0),
|
|
201
|
+
},
|
|
202
|
+
{
|
|
203
|
+
name: 'both sides zero is in band; a sum against a zero census is not',
|
|
204
|
+
expect: 'accept',
|
|
205
|
+
run: () => (reconcile(0, 0, 'b').inBand && !reconcile(5, 0, 'b').inBand ? 0 : 1),
|
|
206
|
+
},
|
|
207
|
+
],
|
|
208
|
+
});
|
|
209
|
+
}
|
|
210
|
+
|
|
211
|
+
export function main(argv: readonly string[] = process.argv.slice(2)): number {
|
|
212
|
+
if (argv.includes('--self-test')) return selfTest();
|
|
213
|
+
const passports = collectPassports();
|
|
214
|
+
try {
|
|
215
|
+
assertScanned({
|
|
216
|
+
gate: 'check_pack_passport_reconciliation',
|
|
217
|
+
scanned: passports.length,
|
|
218
|
+
units: 'pack passport(s)',
|
|
219
|
+
roots: ['src/packs/*/pack.yaml'],
|
|
220
|
+
});
|
|
221
|
+
} catch (err) {
|
|
222
|
+
if (err instanceof DeadScopeError) {
|
|
223
|
+
process.stderr.write(`❌ passport reconciliation: ${err.message}\n`);
|
|
224
|
+
return 2;
|
|
225
|
+
}
|
|
226
|
+
throw err;
|
|
227
|
+
}
|
|
228
|
+
process.stdout.write(`scanned: ${String(passports.length)}\n`);
|
|
229
|
+
|
|
230
|
+
if (argv.includes('--write-ranking')) {
|
|
231
|
+
fs.mkdirSync(path.dirname(RANKING), { recursive: true });
|
|
232
|
+
fs.writeFileSync(RANKING, renderRanking(passports));
|
|
233
|
+
process.stdout.write(`✅ ranking written to ${path.relative(REPO_ROOT, RANKING)}\n`);
|
|
234
|
+
}
|
|
235
|
+
|
|
236
|
+
const rules = censusRuleDir(path.join(REPO_ROOT, 'dist', 'agent-src', 'rules'));
|
|
237
|
+
const catalog = censusSkillsCatalog(path.join(REPO_ROOT, 'dist', 'agent-src', 'skills'));
|
|
238
|
+
const tok = (chars: number): number => Math.round(chars / 4);
|
|
239
|
+
const recs = [
|
|
240
|
+
reconcile(
|
|
241
|
+
passports.reduce((a, p) => a + p.rules_tokens, 0),
|
|
242
|
+
tok(rules.chars),
|
|
243
|
+
'project-scope rules',
|
|
244
|
+
),
|
|
245
|
+
reconcile(
|
|
246
|
+
passports.reduce((a, p) => a + p.catalog_tokens, 0),
|
|
247
|
+
tok(catalog.chars),
|
|
248
|
+
'preloaded skills catalog',
|
|
249
|
+
),
|
|
250
|
+
];
|
|
251
|
+
for (const r of recs) {
|
|
252
|
+
process.stdout.write(
|
|
253
|
+
` ${r.inBand ? '✅' : '❌'} ${r.bucket.padEnd(26)} packs ${String(r.packSum).padStart(7)} · ` +
|
|
254
|
+
`census ${String(r.censusTotal).padStart(7)} · gap ${(r.gapPct * 100).toFixed(1)} %\n`,
|
|
255
|
+
);
|
|
256
|
+
}
|
|
257
|
+
const covered = recs.filter((r) => r.inBand).length;
|
|
258
|
+
process.stdout.write(
|
|
259
|
+
`\ncoverage: packs account for ${String(covered)} of ${String(recs.length)} census bucket(s) within ±${String(
|
|
260
|
+
TOLERANCE * 100,
|
|
261
|
+
)} %.\n` +
|
|
262
|
+
' A gap here is a PROPERTY OF THE TREE, not a defect: pack `dependencies` claim a\n' +
|
|
263
|
+
' small deliberate subset of what ships, so the two numbers are not a reconciliation\n' +
|
|
264
|
+
' pair. Failing on that would be failing by construction, and a gate that is red by\n' +
|
|
265
|
+
' construction gets disabled rather than fixed.\n' +
|
|
266
|
+
' The useful output is the ranking above plus this figure. See\n' +
|
|
267
|
+
' docs/contracts/pack-token-passport.md § Reconciliation for what would make it a gate.\n',
|
|
268
|
+
);
|
|
269
|
+
return 0;
|
|
270
|
+
}
|
|
271
|
+
|
|
272
|
+
if (process.argv[1] !== undefined && process.argv[1].includes('check_pack_passport_reconciliation')) {
|
|
273
|
+
process.exit(main(process.argv.slice(2)));
|
|
274
|
+
}
|
|
@@ -28,6 +28,7 @@ import * as path from 'node:path';
|
|
|
28
28
|
import { fileURLToPath } from 'node:url';
|
|
29
29
|
|
|
30
30
|
import { assertScanned, DeadScopeError } from './_lib/scan_scope.js';
|
|
31
|
+
import { runSelfTest } from './_lib/gate_self_test.js';
|
|
31
32
|
|
|
32
33
|
const REPO_ROOT = path.resolve(fileURLToPath(import.meta.url), '..', '..', '..');
|
|
33
34
|
const BUDGET_PATH = path.join(REPO_ROOT, 'src', 'config', 'pack-size-budget.json');
|
|
@@ -98,6 +99,126 @@ export function skillBytes(files: readonly PackFile[]): { perSkill: Record<strin
|
|
|
98
99
|
}
|
|
99
100
|
|
|
100
101
|
/** Every violation, as human-readable lines. Empty array means green. */
|
|
102
|
+
/**
|
|
103
|
+
* Content classes the published payload is checked for.
|
|
104
|
+
*
|
|
105
|
+
* The gap this closes: before it, the tarball was measured by SIZE and by build
|
|
106
|
+
* correctness, and by nothing that read what the files ARE. A payload can be
|
|
107
|
+
* comfortably under budget and still ship compiled tests, an IDE directory, or
|
|
108
|
+
* a credential-shaped file.
|
|
109
|
+
*
|
|
110
|
+
* `limit: 0` is a hard class — nothing of that shape may ship. A positive limit
|
|
111
|
+
* is a RATCHET: the number was measured, not chosen, and it may only walk down.
|
|
112
|
+
* The distinction is in `measured_in`, which every entry carries, because the
|
|
113
|
+
* same `npm pack` reports a fifth of the source maps in an unbuilt worktree as
|
|
114
|
+
* in a built one. A threshold with no stated tree reads as drift the first time
|
|
115
|
+
* someone runs it from a fresh checkout.
|
|
116
|
+
*/
|
|
117
|
+
interface ContentClass {
|
|
118
|
+
id: string;
|
|
119
|
+
/** Matches a payload path. */
|
|
120
|
+
match: (p: string) => boolean;
|
|
121
|
+
/** 0 = must not ship at all. > 0 = shrink-only ratchet. */
|
|
122
|
+
limit: number;
|
|
123
|
+
/** Which tree the limit was measured in, and when. */
|
|
124
|
+
measured_in: string;
|
|
125
|
+
/**
|
|
126
|
+
* True when the class only exists in a BUILT tree.
|
|
127
|
+
*
|
|
128
|
+
* This is the difference between a check and a false pass. `npm pack
|
|
129
|
+
* --dry-run --ignore-scripts` on a clean checkout — which is the condition
|
|
130
|
+
* `pack-size-budget.json` declares for its own numbers — produces no
|
|
131
|
+
* `dist/cli/**` at all, so a source-map count of 22 sails under a limit of
|
|
132
|
+
* 120 while measuring a fifth of the payload. Reported NOT MEASURABLE
|
|
133
|
+
* instead, on the same principle as the dead-scope assertion above: a check
|
|
134
|
+
* that could not run is not a clean bill.
|
|
135
|
+
*/
|
|
136
|
+
requires_build: boolean;
|
|
137
|
+
why: string;
|
|
138
|
+
}
|
|
139
|
+
|
|
140
|
+
/**
|
|
141
|
+
* Was this payload produced from a built tree?
|
|
142
|
+
*
|
|
143
|
+
* Keyed on the payload itself rather than on the filesystem, so it describes
|
|
144
|
+
* the thing being judged instead of the machine judging it.
|
|
145
|
+
*/
|
|
146
|
+
function payloadIsBuilt(files: readonly PackFile[]): boolean {
|
|
147
|
+
return files.some((f) => f.path.startsWith('dist/cli/'));
|
|
148
|
+
}
|
|
149
|
+
|
|
150
|
+
const CONTENT_CLASSES: readonly ContentClass[] = [
|
|
151
|
+
{
|
|
152
|
+
id: 'compiled-test-artefact',
|
|
153
|
+
requires_build: true,
|
|
154
|
+
match: (p) => /\.(test|spec)\.(js|js\.map|d\.ts)$/.test(p),
|
|
155
|
+
limit: 0,
|
|
156
|
+
measured_in: 'built main checkout, 2026-08-22: 8 `.test.js` + 8 `.test.js.map` were shipping before the `files[]` negations landed',
|
|
157
|
+
why: "a compiled test's output has no consumer-facing purpose; council 2026-08-22 decision (b') strips BOTH the JS and its map, because stripping only the maps left the compiled tests themselves in the tarball",
|
|
158
|
+
},
|
|
159
|
+
{
|
|
160
|
+
id: 'credential-shaped',
|
|
161
|
+
requires_build: false,
|
|
162
|
+
match: (p) => /(^|\/)\.env(\.|$)|\.pem$|(^|\/)id_(rsa|ed25519)$|\.p12$|\.pfx$/.test(p),
|
|
163
|
+
limit: 0,
|
|
164
|
+
measured_in: 'built main checkout, 2026-08-22: 0 present',
|
|
165
|
+
why: 'a clean class with no check is indistinguishable from a class nobody looked at — and this one is worth a canary precisely because it is empty today',
|
|
166
|
+
},
|
|
167
|
+
{
|
|
168
|
+
id: 'ide-metadata',
|
|
169
|
+
requires_build: false,
|
|
170
|
+
match: (p) => /(^|\/)\.(idea|vscode)\//.test(p),
|
|
171
|
+
limit: 0,
|
|
172
|
+
measured_in: 'built main checkout, 2026-08-22: 0 present',
|
|
173
|
+
why: 'same reason as the credential class: empty and unchecked is not the same as empty and verified',
|
|
174
|
+
},
|
|
175
|
+
{
|
|
176
|
+
id: 'source-map',
|
|
177
|
+
requires_build: true,
|
|
178
|
+
match: (p) => p.endsWith('.js.map'),
|
|
179
|
+
limit: 120,
|
|
180
|
+
measured_in: 'built worktree, 2026-08-22, AFTER the compiled-test negations: 120 product maps (128 total minus 8 test maps)',
|
|
181
|
+
why: "a product source map is a consumer debugging affordance — proper line mapping and stepping over shipped JS — so council 2026-08-22 kept the 119/120 product maps and refused a blanket zero. Recorded as a PROVISIONAL measured ratchet, not an architectural constant: nothing here establishes that any consumer debugs the shipped JS, and nothing establishes they do not",
|
|
182
|
+
},
|
|
183
|
+
];
|
|
184
|
+
|
|
185
|
+
/**
|
|
186
|
+
* Count each content class in the payload and report over-limit classes.
|
|
187
|
+
*
|
|
188
|
+
* Returns one error string per violating class, with the offending paths — the
|
|
189
|
+
* count alone tells a reader a rule broke and not which file broke it.
|
|
190
|
+
*/
|
|
191
|
+
export function classifyPayload(files: readonly PackFile[]): string[] {
|
|
192
|
+
const out: string[] = [];
|
|
193
|
+
const built = payloadIsBuilt(files);
|
|
194
|
+
for (const c of CONTENT_CLASSES) {
|
|
195
|
+
if (c.requires_build && !built) continue;
|
|
196
|
+
const hits = files.filter((f) => c.match(f.path)).map((f) => f.path);
|
|
197
|
+
if (hits.length <= c.limit) continue;
|
|
198
|
+
const shown = hits.slice(0, 8);
|
|
199
|
+
out.push(
|
|
200
|
+
`content class \`${c.id}\`: ${String(hits.length)} entr${hits.length === 1 ? 'y' : 'ies'} ` +
|
|
201
|
+
`exceeds its limit of ${String(c.limit)} (measured in: ${c.measured_in}) — ` +
|
|
202
|
+
`${shown.join(', ')}${hits.length > shown.length ? `, +${String(hits.length - shown.length)} more` : ''}. ` +
|
|
203
|
+
`Why this class is checked: ${c.why}`,
|
|
204
|
+
);
|
|
205
|
+
}
|
|
206
|
+
return out;
|
|
207
|
+
}
|
|
208
|
+
|
|
209
|
+
/** Per-class counts, for the green-path report. */
|
|
210
|
+
export function payloadClassCounts(
|
|
211
|
+
files: readonly PackFile[],
|
|
212
|
+
): Array<{ id: string; count: number; limit: number; measurable: boolean }> {
|
|
213
|
+
const built = payloadIsBuilt(files);
|
|
214
|
+
return CONTENT_CLASSES.map((c) => ({
|
|
215
|
+
id: c.id,
|
|
216
|
+
count: files.filter((f) => c.match(f.path)).length,
|
|
217
|
+
limit: c.limit,
|
|
218
|
+
measurable: built || !c.requires_build,
|
|
219
|
+
}));
|
|
220
|
+
}
|
|
221
|
+
|
|
101
222
|
export function evaluate(budget: PackSizeBudget, pack: PackResult): string[] {
|
|
102
223
|
const errors: string[] = [];
|
|
103
224
|
|
|
@@ -163,7 +284,103 @@ function readPack(argv: readonly string[]): PackResult {
|
|
|
163
284
|
return parsePackJson(stdout);
|
|
164
285
|
}
|
|
165
286
|
|
|
287
|
+
/**
|
|
288
|
+
* Per-class discrimination, proven rather than asserted.
|
|
289
|
+
*
|
|
290
|
+
* WHY THIS EXISTS AND THE CANARY DOES NOT COVER IT. `gate-coverage.yml` carries
|
|
291
|
+
* `canary?: CanarySpec` — exactly ONE recipe per gate id, not a list (see
|
|
292
|
+
* `check_gate_coverage.ts`'s entry shape). The roadmap step behind this asked
|
|
293
|
+
* for "one canary per class", which the register cannot express. Two of the four
|
|
294
|
+
* classes cannot be planted through it at all: `compiled-test-artefact` is
|
|
295
|
+
* excluded from the payload by the `files[]` negations that fixed it, so a plant
|
|
296
|
+
* never ships and never reds, and `source-map` needs 121 files in a BUILT tree.
|
|
297
|
+
*
|
|
298
|
+
* So the register gets the one canary it can hold — the credential-shaped plant,
|
|
299
|
+
* chosen because that class is empty today and an empty unchecked class is
|
|
300
|
+
* indistinguishable from one nobody looked at — and every class is proven red
|
|
301
|
+
* here, over a synthetic payload, deterministically and with no pack run.
|
|
302
|
+
*
|
|
303
|
+
* A test file was the alternative and is weaker: the non-adopter ratchet in
|
|
304
|
+
* `check_gate_coverage` requires a NEW registered gate to carry a `--self-test`
|
|
305
|
+
* or an exemption, and a self-test drives the gate's own decision function
|
|
306
|
+
* rather than a copy of it.
|
|
307
|
+
*/
|
|
308
|
+
function packFiles(...paths: readonly string[]): PackFile[] {
|
|
309
|
+
return paths.map((path) => ({ path, size: 1 }));
|
|
310
|
+
}
|
|
311
|
+
|
|
312
|
+
/** A payload that looks built, so `requires_build` classes do not abstain. */
|
|
313
|
+
function builtPayload(...paths: readonly string[]): PackFile[] {
|
|
314
|
+
return packFiles('dist/cli/agent-config.js', ...paths);
|
|
315
|
+
}
|
|
316
|
+
|
|
317
|
+
function selfTest(): number {
|
|
318
|
+
// NOT under `dist/cli/` — that prefix is exactly what `payloadIsBuilt`
|
|
319
|
+
// keys on, so maps placed there would make an "unbuilt" fixture built and
|
|
320
|
+
// the abstain case would test nothing. Found by the case failing, which is
|
|
321
|
+
// the self-test doing its job on its own fixtures.
|
|
322
|
+
const maps = (n: number): string[] =>
|
|
323
|
+
Array.from({ length: n }, (_, i) => `dist/hooks/chunk-${String(i)}.js.map`);
|
|
324
|
+
return runSelfTest({
|
|
325
|
+
gate: 'check_pack_size',
|
|
326
|
+
minCases: 9,
|
|
327
|
+
minRejectCases: 5,
|
|
328
|
+
cases: [
|
|
329
|
+
{
|
|
330
|
+
name: 'credential-shaped: a .pem in the payload is refused',
|
|
331
|
+
expect: 'reject',
|
|
332
|
+
run: () => (classifyPayload(packFiles('src/scripts/x.pem')).length > 0 ? 1 : 0),
|
|
333
|
+
},
|
|
334
|
+
{
|
|
335
|
+
name: 'credential-shaped: a dotted .env form is refused',
|
|
336
|
+
expect: 'reject',
|
|
337
|
+
run: () => (classifyPayload(packFiles('src/config/.env.production')).length > 0 ? 1 : 0),
|
|
338
|
+
},
|
|
339
|
+
{
|
|
340
|
+
name: 'ide-metadata: a .vscode entry is refused',
|
|
341
|
+
expect: 'reject',
|
|
342
|
+
run: () => (classifyPayload(packFiles('src/scripts/.vscode/settings.json')).length > 0 ? 1 : 0),
|
|
343
|
+
},
|
|
344
|
+
{
|
|
345
|
+
name: 'compiled-test-artefact: a shipped .test.js is refused',
|
|
346
|
+
expect: 'reject',
|
|
347
|
+
run: () => (classifyPayload(builtPayload('dist/cli/a.test.js')).length > 0 ? 1 : 0),
|
|
348
|
+
},
|
|
349
|
+
{
|
|
350
|
+
name: 'source-map: one over the measured ratchet is refused',
|
|
351
|
+
expect: 'reject',
|
|
352
|
+
run: () => (classifyPayload(builtPayload(...maps(121))).length > 0 ? 1 : 0),
|
|
353
|
+
},
|
|
354
|
+
{
|
|
355
|
+
name: 'source-map: the ratchet boundary itself passes',
|
|
356
|
+
expect: 'accept',
|
|
357
|
+
run: () => (classifyPayload(builtPayload(...maps(120))).length > 0 ? 1 : 0),
|
|
358
|
+
},
|
|
359
|
+
{
|
|
360
|
+
name: 'a plain payload passes — the classes are not firing on everything',
|
|
361
|
+
expect: 'accept',
|
|
362
|
+
run: () => (classifyPayload(packFiles('src/scripts/x.ts', 'README.md')).length > 0 ? 1 : 0),
|
|
363
|
+
},
|
|
364
|
+
{
|
|
365
|
+
// The false pass 1.2 closed. An unbuilt payload carries a fifth
|
|
366
|
+
// of the source maps, so a build-gated class must ABSTAIN there
|
|
367
|
+
// rather than report clean — and must not report a failure
|
|
368
|
+
// either, or every clean-checkout run would red.
|
|
369
|
+
name: 'unbuilt payload: build-gated classes abstain, not pass and not fail',
|
|
370
|
+
expect: 'accept',
|
|
371
|
+
run: () => (classifyPayload(packFiles(...maps(121))).length > 0 ? 1 : 0),
|
|
372
|
+
},
|
|
373
|
+
{
|
|
374
|
+
name: 'unbuilt payload: a build-INDEPENDENT class still fires',
|
|
375
|
+
expect: 'reject',
|
|
376
|
+
run: () => (classifyPayload(packFiles('src/scripts/id_rsa')).length > 0 ? 1 : 0),
|
|
377
|
+
},
|
|
378
|
+
],
|
|
379
|
+
});
|
|
380
|
+
}
|
|
381
|
+
|
|
166
382
|
export function main(argv: readonly string[] = process.argv.slice(2)): number {
|
|
383
|
+
if (argv.includes('--self-test')) return selfTest();
|
|
167
384
|
let budget: PackSizeBudget;
|
|
168
385
|
let pack: PackResult;
|
|
169
386
|
try {
|
|
@@ -199,7 +416,14 @@ export function main(argv: readonly string[] = process.argv.slice(2)): number {
|
|
|
199
416
|
throw err;
|
|
200
417
|
}
|
|
201
418
|
|
|
202
|
-
|
|
419
|
+
// The coverage register requires an enforced gate to report what it
|
|
420
|
+
// inspected, and it is the same reason `assertScanned` runs above: a verdict
|
|
421
|
+
// with no denominator cannot be told apart from a verdict over nothing.
|
|
422
|
+
// Printed on EVERY path, including the failing one, so a reader of a red run
|
|
423
|
+
// still knows the population it was red over.
|
|
424
|
+
process.stdout.write(`scanned: ${String(pack.files.length)}\n`);
|
|
425
|
+
|
|
426
|
+
const errors = [...evaluate(budget, pack), ...classifyPayload(pack.files)];
|
|
203
427
|
const { perSkill, total } = skillBytes(pack.files);
|
|
204
428
|
if (argv.includes('--json')) {
|
|
205
429
|
process.stdout.write(
|
|
@@ -216,6 +440,15 @@ export function main(argv: readonly string[] = process.argv.slice(2)): number {
|
|
|
216
440
|
`${Object.keys(perSkill).length} skills, largest share ` +
|
|
217
441
|
`${Math.max(...Object.values(perSkill).map((b) => (b / total) * 100)).toFixed(2)}%\n`,
|
|
218
442
|
);
|
|
443
|
+
// Printed on the GREEN path on purpose. A class that is clean and silent is
|
|
444
|
+
// indistinguishable from a class nobody checks, which is the whole reason
|
|
445
|
+
// two of these four are limit-0 over an empty set today.
|
|
446
|
+
for (const c of payloadClassCounts(pack.files)) {
|
|
447
|
+
const verdict = c.measurable
|
|
448
|
+
? `${String(c.count)} / ${c.limit === 0 ? 'must be 0' : `${String(c.limit)} max`}`
|
|
449
|
+
: 'NOT MEASURABLE — this payload has no dist/cli/**, so the tree is unbuilt and the class is out of view';
|
|
450
|
+
process.stdout.write(` content class ${c.id}: ${verdict}\n`);
|
|
451
|
+
}
|
|
219
452
|
return 0;
|
|
220
453
|
}
|
|
221
454
|
|