code-foundry 1.13.0 → 1.14.0

package/CHANGELOG.md

@@ -1,5 +1,12 @@
 # Changelog
 
+## [1.14.0](https://github.com/0xPlayerOne/code-foundry/compare/v1.13.0...v1.14.0) (2026-09-09)
+
+
+### Features
+
+* **cli:** add shared validation plans and agent check evidence ([#538](https://github.com/0xPlayerOne/code-foundry/issues/538)) ([95af9ed](https://github.com/0xPlayerOne/code-foundry/commit/95af9ed42682ff3ed1dc74609ea4628d0c95a799))
+
 ## [1.13.0](https://github.com/0xPlayerOne/code-foundry/compare/v1.12.0...v1.13.0) (2026-09-09)
 
 

package/docs/PERFORMANCE.md

@@ -16,8 +16,8 @@ belong to the run that produced them, not the source tree.
 | Runtime dependencies                |      0 | Keep the installed CLI dependency-free               |
 | Development dependencies            |      4 | Prevent unreviewed toolchain growth                  |
 | Packed artifact                     | 210 kB | Bound registry transfer and install cost             |
-| Unpacked artifact                   | 800 kB | Bound installed footprint                            |
-| Packed files                        |     90 | Detect accidental release contents                   |
+| Unpacked artifact                   | 820 kB | Bound installed footprint                            |
+| Packed files                        |     92 | Detect accidental release contents                   |
 
 The performance workflow disables build-cache reads and writes for this task.
 That makes timing comparisons independent of a warm protected-branch cache and

package/docs/README.md

@@ -14,6 +14,7 @@ its own names, environments, and deployment details.
 - [Caching and remote caching](CACHING.md)
 - [Performance budgets and baselines](PERFORMANCE.md)
 - [Required capabilities and task evidence](required-capabilities.md)
+- [Agent-facing validation commands](agent-validation.md)
 - [Declarative fleet inventory and staged rollouts](fleet-rollouts.md)
 
 ## Repository-specific documentation

package/docs/agent-validation.md

@@ -0,0 +1,94 @@
+# Agent-facing validation commands
+
+The public CLI now exposes a stable plan/check interface using the same discovery,
+required-capability policy, ecosystem executor, and task receipts as reusable CI.
+
+```sh
+code-foundry plan --changed --base origin/main --json
+code-foundry check --tier fast --json
+code-foundry check --tier audit --json
+```
+
+Both commands accept `--target PATH`. Use the Code Foundry version pinned by the
+repository, not an unreviewed floating installation. `plan` executes discovery
+only: it does not install dependencies, run project checks, modify source files,
+or emit skip-receipt files. Required entrypoints are validated across the complete
+task set, including tasks deferred from the selected local tier.
+
+## Tiers and change awareness
+
+`fast` selects formatting, linting, type checking, build, unit tests, and performance.
+`audit` adds integration, E2E, and smoke tests. Ecosystem commands and applicability
+come from the shared runtime; the CLI does not invent alternate test commands.
+Repository-owned scripts remain authoritative. Missing required tasks fail
+planning instead of returning a misleading successful subset.
+
+`--changed` includes paths changed from the selected base, staged changes,
+unstaged changes, and untracked files. Committed paths use the merge-base with
+that ref, so an independently advanced base branch does not make base-only
+changes look like head changes. `--base` defaults to HEAD and requires `--changed`.
+Invalid or unavailable base revisions fail rather than being treated as no changes.
+Paths use Git's NUL-separated format, preserving whitespace and unusual filenames.
+Generated `.code-foundry/` evidence is excluded from change annotations. Change awareness is **annotation-only**: without a verified dependency
+graph it does not skip required tasks or assume a documentation change cannot
+affect a custom command.
+
+Local tiers are not a replacement for the complete GitHub Validation / Gate.
+Security scans, CodeQL, release-diff policy where applicable, protected-environment
+checks, and required reviews remain separate. Every plan/result explicitly marks
+remote validation as required. A fast result also names deferred local tasks.
+A local audit result is not permission to mark a PR ready or merge it.
+
+## Execution and evidence
+
+Install the repository's locked dependencies through its normal setup process
+before running checks. This CLI does not add an implicit install step or grant
+credentials. It runs trusted repository scripts in the existing environment;
+those scripts are not sandboxed and retain their normal tool/network behavior.
+The formatter/runtime's existing semantics are unchanged: a configured command
+that rewrites files still rewrites files. Prefer check-only scripts in CI.
+
+Task stdout/stderr goes to stderr, keeping `--json` stdout parseable. Each selected
+applicable task runs through the public runtime. A zero exit without a fresh,
+matching, successful task receipt fails. A failed task blocks remaining applicable
+tasks; optional inapplicable tasks are reported as skipped. An entirely skipped
+check is labeled `skipped`, never `passed`.
+
+A run stores `.code-foundry/agent-results/check-*/summary.json` and snapshots of
+its task receipts. The result includes the source commit, dirty-tree indication,
+selected/deferred tasks, reasons, exit codes, signals, receipt paths, and repository
+artifact paths such as coverage or browser evidence. HEAD identifies the base
+commit; a dirty working tree is not a cryptographically identified immutable
+source snapshot. Reports are local evidence, not signed provenance.
+
+The aggregate and task receipts are retained per run so later checks do not
+overwrite the earlier JSON evidence. Artifact paths inside receipts point to
+repository-owned files and may be overwritten by subsequent test runs; copy them
+into a CI artifact for long-term retention. Do not upload the whole hidden
+repository tree or secrets. The CLI neither captures environment variables into
+reports nor grants publish/merge/deploy access.
+
+An exclusive local `agent-results/active.lock` prevents overlapping agent checks
+from confusing receipts. After a killed process, inspect the checkout and confirm
+there is no active check before manually removing a stale lock. Normal completion
+and failures clean the lock. This is a local CLI lock, not a distributed lock
+against every other tool writing into the repository.
+
+`--timeout SECONDS` bounds each check subprocess (default 600, range 1–3600).
+A timeout or missing receipt fails. Project scripts remain responsible for
+cleaning up their own servers and subprocesses. Plans do not accept an execution
+timeout. Unknown, duplicate, or ignored arguments fail rather than silently alter
+validation behavior.
+
+## Integration and tests
+
+This change depends on the required-capabilities/task-evidence runtime change.
+It deliberately imports shared policy instead of maintaining a competing copy.
+Merge that prerequisite before this CLI change and retain `runtime-core.mjs` when
+vendoring the runtime.
+
+Run `node --test test/agent-check.test.mjs`. Tests exercise real Git change
+inspection and the actual policy/evidence wrapper with a deterministic ecosystem
+executor, including read-only planning, required deferred tasks, failed commands,
+stale evidence, receipt snapshots, locking, and separation of JSON from task logs.
+Full ecosystem dependencies and GitHub-side checks still need their own CI run.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "code-foundry",
-  "version": "1.13.0",
+  "version": "1.14.0",
   "description": "A fast, language-aware repository factory for agent-ready workflows, testing, security, and release automation.",
   "homepage": "https://github.com/0xPlayerOne/code-foundry#readme",
   "bugs": {

package/src/cli.mjs

@@ -16,6 +16,8 @@ Usage:
   npx code-foundry init [--target PATH]
   npx code-foundry sync [--target PATH]
   npx code-foundry doctor [--target PATH]
+  npx code-foundry plan [--target PATH] [--tier fast|audit] [--changed] [--base REF] [--json]
+  npx code-foundry check [--target PATH] [--tier fast|audit] [--changed] [--base REF] [--timeout SECONDS] [--json]
   npx code-foundry ci pause|resume|status [--target PATH]
   npx code-foundry release reconcile [--github] [--base BRANCH] [--head BRANCH]
   npx code-foundry release hook --tag TAG --workflow WORKFLOW
@@ -124,6 +126,11 @@ function parseArgs(argv) {
 }
 
 async function main() {
+  if (['plan', 'check'].includes(process.argv[2])) {
+    const { agentCommand } = await import('./commands/agent-check.mjs')
+    process.exitCode = agentCommand(process.argv.slice(2))
+    return
+  }
   const { command, options } = parseArgs(process.argv.slice(2))
   const target = resolve(options.target)
 

package/src/commands/agent-check.mjs

@@ -0,0 +1,309 @@
+// @ts-check
+
+import {
+  existsSync,
+  lstatSync,
+  mkdirSync,
+  mkdtempSync,
+  readFileSync,
+  rmSync,
+  writeFileSync,
+} from 'node:fs'
+import { join, resolve } from 'node:path'
+import { spawnSync } from 'node:child_process'
+import { fileURLToPath } from 'node:url'
+import { taskProfile } from '../runtime.mjs'
+import { evidencePath, fingerprint, TASKS } from '../lib/task-policy.mjs'
+
+const runtime = fileURLToPath(new URL('../runtime.mjs', import.meta.url))
+const fastTasks = new Set(['format', 'lint', 'type_check', 'build', 'unit', 'performance'])
+export const agentUsage = `code-foundry plan [--target PATH] [--tier fast|audit] [--changed] [--base REF] [--json]
+code-foundry check [--target PATH] [--tier fast|audit] [--changed] [--base REF] [--timeout SECONDS] [--json]
+
+plan discovers tasks without executing them. check invokes the shared CI runtime.
+--changed annotates affected paths; it never prunes required or custom checks.
+fast includes CI tasks, unit tests, and performance; audit adds integration/E2E/smoke.
+Neither command substitutes for GitHub Security, CodeQL, release policy, or review.
+`
+
+/** @typedef {{command: 'plan'|'check', target: string, tier: 'fast'|'audit', changed: boolean, base: string, json: boolean, timeout: number}} AgentOptions */
+/** @typedef {{runtime?: string, core?: string}} RuntimePaths */
+
+/** @param {string[]} argv @returns {AgentOptions} */
+export function parseAgentArgs(argv) {
+  const [command, ...args] = argv
+  if (command !== 'plan' && command !== 'check') throw new Error('Expected plan or check')
+  /** @type {AgentOptions} */
+  const options = {
+    command,
+    target: process.cwd(),
+    tier: 'fast',
+    changed: false,
+    base: 'HEAD',
+    json: false,
+    timeout: 600000,
+  }
+  const seen = new Set()
+  while (args.length) {
+    const key = args.shift()
+    if (!key || seen.has(key)) throw new Error(`Duplicate or missing option: ${key}`)
+    seen.add(key)
+    if (key === '--json') options.json = true
+    else if (key === '--changed') options.changed = true
+    else {
+      if (!['--target', '--tier', '--base', '--timeout'].includes(key))
+        throw new Error(`Unknown option: ${key}`)
+      const value = args.shift()
+      if (!value || value.startsWith('-')) throw new Error(`Missing value for ${key}`)
+      if (key === '--target') options.target = resolve(value)
+      if (key === '--base') options.base = value
+      if (key === '--tier') {
+        if (!['fast', 'audit'].includes(value)) throw new Error('tier must be fast or audit')
+        options.tier = /** @type {'fast'|'audit'} */ (value)
+      }
+      if (key === '--timeout') {
+        const seconds = Number(value)
+        if (!Number.isSafeInteger(seconds) || seconds < 1 || seconds > 3600)
+          throw new Error('timeout must be 1–3600 whole seconds')
+        options.timeout = seconds * 1000
+      }
+    }
+  }
+  if (seen.has('--base') && !options.changed) throw new Error('--base requires --changed')
+  if (command === 'plan' && seen.has('--timeout'))
+    throw new Error('--timeout applies only to check execution')
+  return options
+}
+
+/** @param {string} root @param {string[]} args */
+function git(root, args) {
+  const result = spawnSync('git', args, {
+    cwd: root,
+    encoding: 'utf8',
+    timeout: 10000,
+    maxBuffer: 8 * 1024 * 1024,
+  })
+  if (result.error || result.status !== 0)
+    throw new Error(`Unable to inspect repository: git ${args[0]}`)
+  return result.stdout
+}
+
+/** @param {string} value */
+function hasInvalidBaseRefCharacter(value) {
+  return [...value].some((character) => {
+    const code = character.codePointAt(0) ?? 0
+    return code <= 0x20 || code === 0x7f
+  })
+}
+
+/** @param {AgentOptions} options */
+export function sourceContext(options) {
+  const sourceSha = git(options.target, ['rev-parse', '--verify', 'HEAD']).trim()
+  if (!/^[0-9a-f]{40}$/i.test(sourceSha))
+    throw new Error('Repository HEAD must resolve to a commit')
+  const dirty = git(options.target, ['status', '--porcelain=v1', '-z'])
+    .split('\0')
+    .filter(Boolean)
+    .some((line) => !line.slice(3).startsWith('.code-foundry/'))
+  let baseSha = null
+  /** @type {string[]} */
+  let changedFiles = []
+  if (options.changed) {
+    if (options.base.startsWith('-') || hasInvalidBaseRefCharacter(options.base))
+      throw new Error('Invalid base ref')
+    baseSha = git(options.target, ['rev-parse', '--verify', `${options.base}^{commit}`]).trim()
+    const lists = [
+      git(options.target, ['diff', '--name-only', '-z', `${baseSha}...HEAD`, '--']),
+      git(options.target, ['diff', '--cached', '--name-only', '-z', '--']),
+      git(options.target, ['diff', '--name-only', '-z', '--']),
+      git(options.target, ['ls-files', '--others', '--exclude-standard', '-z']),
+    ]
+    changedFiles = [
+      ...new Set(
+        lists
+          .flatMap((value) => value.split('\0'))
+          .filter((file) => file && !file.startsWith('.code-foundry/'))
+      ),
+    ]
+    // This freshly created array is local to the plan.
+    // oxlint-disable-next-line unicorn/no-array-sort
+    changedFiles.sort()
+  }
+  return { sourceSha, dirty, baseSha, changedFiles }
+}
+
+/** @param {AgentOptions} options @param {RuntimePaths} [paths] */
+export function planChecks(options, paths = {}) {
+  const source = sourceContext(options)
+  // Direct discovery shares CI policy, but unlike ci task_profile it writes no skip receipts.
+  const tasks = TASKS.map((task) => ({
+    ...taskProfile(options.target, task, paths.core),
+    selected: options.tier === 'audit' || fastTasks.has(task),
+    argv: [process.execPath, paths.runtime ?? runtime, 'ci', task],
+  }))
+  return {
+    schemaVersion: 1,
+    kind: 'code-foundry-validation-plan',
+    status: 'planned',
+    tier: options.tier,
+    ...source,
+    tasks,
+    changePolicy: 'annotation-only; no tasks pruned without a proven dependency graph',
+    remoteValidationRequired: true,
+    deferredRemoteChecks: [
+      'Security',
+      'CodeQL',
+      'release policy when applicable',
+      'required reviews',
+    ],
+  }
+}
+
+/** @param {string} root */
+function outputDirectory(root) {
+  let directory = resolve(root)
+  for (const component of ['.code-foundry', 'agent-results']) {
+    directory = join(directory, component)
+    if (!existsSync(directory)) mkdirSync(directory)
+    if (lstatSync(directory).isSymbolicLink() || !lstatSync(directory).isDirectory())
+      throw new Error('Agent evidence directory must be a real directory inside the repository')
+  }
+  return directory
+}
+
+/** @param {AgentOptions} options @param {RuntimePaths} [paths] */
+export function checkRepository(options, paths = {}) {
+  const plan = planChecks(options, paths)
+  const directory = outputDirectory(options.target)
+  const lock = join(directory, 'active.lock')
+  try {
+    mkdirSync(lock)
+  } catch {
+    throw new Error(
+      'Another agent check may be running; inspect agent-results/active.lock before recovery'
+    )
+  }
+  let runDirectory
+  try {
+    runDirectory = mkdtempSync(join(directory, 'check-'))
+  } catch (error) {
+    try {
+      rmSync(lock, { recursive: true, force: true })
+    } catch {
+      // Preserve the original temporary-directory error.
+    }
+    throw error
+  }
+  const reportFile = join(runDirectory, 'summary.json')
+  /** @type {Record<string, any>} */
+  const report = {
+    ...plan,
+    kind: 'code-foundry-validation-result',
+    status: 'failed',
+    startedAt: new Date().toISOString(),
+    completedAt: null,
+    tasks: [],
+    report: reportFile,
+    deferredTasks: plan.tasks.filter((task) => !task.selected).map((task) => task.task),
+  }
+  let failed = false
+  try {
+    for (const item of plan.tasks.filter((task) => task.selected)) {
+      if (!item.applicable || failed) {
+        report.tasks.push({
+          ...item,
+          status: failed && item.applicable ? 'blocked' : 'skipped',
+          reason: failed && item.applicable ? 'an earlier task failed' : item.reason,
+        })
+        continue
+      }
+      const receiptFile = evidencePath(options.target, `.code-foundry/results/${item.task}.json`)
+      const before = fingerprint(receiptFile)
+      const result = spawnSync(item.argv[0], item.argv.slice(1), {
+        cwd: options.target,
+        timeout: options.timeout,
+        // Keep stdout exclusively available for the public JSON contract.
+        stdio: ['ignore', 2, 2],
+        env: { ...process.env, GITHUB_OUTPUT: '', GITHUB_STEP_SUMMARY: '' },
+      })
+      let receipt = null
+      let reason = result.error
+        ? `Runtime execution failed: ${/** @type {NodeJS.ErrnoException} */ (result.error).code ?? result.error.name}`
+        : ''
+      if (fingerprint(receiptFile) !== before && existsSync(receiptFile)) {
+        try {
+          receipt = JSON.parse(readFileSync(receiptFile, 'utf8'))
+          if (
+            receipt.kind !== 'code-foundry-task-result' ||
+            receipt.schemaVersion !== 1 ||
+            receipt.task !== item.task ||
+            receipt.sourceSha !== plan.sourceSha
+          )
+            throw new Error('Receipt identity does not match the planned task/source')
+        } catch (error) {
+          reason = error instanceof Error ? error.message : String(error)
+          receipt = null
+        }
+      }
+      const passed = result.status === 0 && receipt?.status === 'passed' && !reason
+      const snapshot = receipt ? join(runDirectory, `${item.task}.json`) : null
+      if (snapshot) writeFileSync(snapshot, `${JSON.stringify(receipt, null, 2)}\n`)
+      failed = !passed
+      report.tasks.push({
+        ...item,
+        status: passed ? 'passed' : 'failed',
+        exitCode: result.status,
+        signal: result.signal,
+        reason: reason || receipt?.reason || 'missing fresh successful task evidence',
+        receipt: snapshot,
+        artifacts: receipt?.artifacts ?? [],
+      })
+    }
+    report.status = failed
+      ? 'failed'
+      : report.tasks.some((/** @type {any} */ task) => task.status === 'passed')
+        ? 'passed'
+        : 'skipped'
+    return report
+  } finally {
+    report.completedAt = new Date().toISOString()
+    try {
+      writeFileSync(reportFile, `${JSON.stringify(report, null, 2)}\n`)
+    } finally {
+      rmSync(lock, { recursive: true })
+    }
+  }
+}
+
+/** @param {string[]} argv */
+export function agentCommand(argv) {
+  const jsonOutput = argv.includes('--json')
+  if (argv.includes('--help') || argv.includes('-h')) {
+    console.log(agentUsage)
+    return 0
+  }
+  try {
+    const options = parseAgentArgs(argv)
+    const result = options.command === 'plan' ? planChecks(options) : checkRepository(options)
+    if (options.json) console.log(JSON.stringify(result, null, 2))
+    else {
+      console.log(`${options.command}: ${result.status} (${options.tier}, ${result.sourceSha})`)
+      for (const item of result.tasks)
+        console.log(
+          `${item.task}: ${item.status ?? (item.selected ? (item.applicable ? 'planned' : 'skipped') : 'deferred')} — ${item.reason}`
+        )
+      console.log('GitHub security checks and review requirements remain separate.')
+    }
+    return result.status === 'failed' ? 1 : 0
+  } catch (error) {
+    const result = {
+      schemaVersion: 1,
+      kind: 'code-foundry-validation-error',
+      status: 'failed',
+      reason: error instanceof Error ? error.message : String(error),
+    }
+    if (jsonOutput) console.log(JSON.stringify(result, null, 2))
+    else console.error(result.reason)
+    return 1
+  }
+}