forgedock 1.27.0 → 1.28.0

package/bin/engine.mjs

@@ -945,7 +945,7 @@ export async function runIssue(opts) {
         // "needs-human" — this is the engine/tool breaking, not a genuine
         // human-judgment block (see #2244/#2261). Any other thrown error is
         // a true unexpected crash and keeps propagating unchanged.
-        if (e.code === "NO_API_KEY" || e.code === "NO_SDK" || e.code === "CLI_BACKEND_FAILED" || e.code === "CLI_MAX_TURNS" || isDiskQuotaError(e)) {
+        if (e.code === "NO_API_KEY" || e.code === "NO_SDK" || e.code === "CLI_BACKEND_FAILED" || e.code === "CLI_BACKEND_UNAVAILABLE" || e.code === "CLI_MAX_TURNS" || isDiskQuotaError(e)) {
           // forge#2241: when the runner attached a session-limit reset time
           // (bin/runner.mjs's extractSessionLimitResetTime(), only ever set
           // for a genuine session-limit CLI_BACKEND_FAILED — never
@@ -1055,6 +1055,8 @@ export async function runIssue(opts) {
       usage: outcome.usage ?? null, last_attempt_usage: outcome.last_attempt_usage ?? null,
       // forge#3589: model + turn count per phase (null when the backend does not report them).
       model: outcome.model ?? null, turns: outcome.turns ?? null,
+      // forge#4257: the execution backend that ran the phase (null when unreported).
+      backend: outcome.backend ?? null,
       ...(commitReason ? { terminalReason: commitReason, code: commitCode } : {}),
     });
     state = deriveState(readLog(dir, issue));
@@ -1282,6 +1284,7 @@ async function runPhaseWithRetry(phase, state, ctx) {
       event: "PHASE_ATTEMPT_USAGE", phase: phase.id, attempt,
       usage: usage ?? null, duration_ms: Math.round(performance.now() - t0),
       cost_usd: extra?.cost_usd ?? null, model: extra?.model ?? null, turns: extra?.turns ?? null,
+      backend: extra?.backend ?? null,
     };
     appendEvent(dir, issue, rec);
     try { await emitTelemetry?.({ ...rec, issue }); } catch { /* telemetry never fails a phase */ }
@@ -1348,7 +1351,7 @@ async function runPhaseWithRetry(phase, state, ctx) {
         attempt--;
         continue;
       }
-      if (e.code === "NO_API_KEY" || e.code === "NO_SDK" || e.code === "CLI_BACKEND_FAILED") throw e;
+      if (e.code === "NO_API_KEY" || e.code === "NO_SDK" || e.code === "CLI_BACKEND_FAILED" || e.code === "CLI_BACKEND_UNAVAILABLE") throw e;
       // forge#2377: no `usage` field here — the runner threw, so no result
       // (and therefore no usage data) was ever produced for this attempt.
       // Do not fabricate a value; omitting the field (rather than a stale
@@ -1367,10 +1370,11 @@ async function runPhaseWithRetry(phase, state, ctx) {
     // forge#3589: model/turns recorded alongside usage (additive; null when absent).
     const lastModel = result?.cliModel ?? result?.model ?? null;
     const lastTurns = result?.turns ?? null;
+    const lastBackend = typeof result?.backend === "string" ? result.backend : null; // forge#4257
     addUsage(lastUsage);
-    await recordAttempt(attempt, lastUsage, t0, { cost_usd: result?.costUsd ?? null, model: lastModel, turns: lastTurns });
+    await recordAttempt(attempt, lastUsage, t0, { cost_usd: result?.costUsd ?? null, model: lastModel, turns: lastTurns, backend: lastBackend });
     const outcome = await phase.detectOutcome(state, io, result);
-    if (outcome.status === "committed" || outcome.status === "blocked") return { ...outcome, usage: cumulativeUsage, last_attempt_usage: lastUsage, model: lastModel, turns: lastTurns };
+    if (outcome.status === "committed" || outcome.status === "blocked") return { ...outcome, usage: cumulativeUsage, last_attempt_usage: lastUsage, model: lastModel, turns: lastTurns, backend: lastBackend };
     appendEvent(dir, issue, { event: "PHASE_FAILED", phase: phase.id, attempt, reason: outcome.detail, code: normalizeCode(outcome.code), maxAttempts, usage: lastUsage, model: lastModel, turns: lastTurns });
     // forge#2176: a phase's detectOutcome can mark a failure as a known,
     // state-derived fixed point — re-running the phase's runner is

package/bin/forgedock.mjs

@@ -3231,6 +3231,26 @@ async function doctor(fix = false) {
     );
   }
 
+  // ── Check 13: Resource hygiene (#4183) ──────────────────────────────────────
+  // Owned leftovers (stale lane ledgers, forgedock-labelled docker, lean-*/agent-* worktrees, lane branches,
+  // ~/.forge/tmp scratch) with counts and sizes; unowned stale ones are only reported. --fix runs the reaper with the
+  // forge.yaml hygiene TTLs. Never a failure: a leftover slows the host, it does not break the install.
+  {
+    try {
+      const gc = await import("./gc.mjs");
+      const probes = gc.makeGcProbes({ cwd: process.cwd() });
+      const h = gc.readHygiene(process.cwd());
+      const plan = await gc.planGc({ probes, ttlMs: gc.hygieneTtls(h), deleteMergedBranches: h.delete_merged_branches });
+      const applied = fix ? await gc.applyGc(plan, probes) : null;
+      if (applied && applied.reclaimed.length > 0) fixesApplied++;
+      const r = gc.hygieneCheck(plan, { applied });
+      if (r.status === "pass") pass("Resource hygiene", r.detail);
+      else warn("Resource hygiene", r.detail);
+    } catch (e) {
+      warn("Resource hygiene", `could not inspect ForgeDock leftovers: ${String(e?.message ?? e).slice(0, 200)}`);
+    }
+  }
+
   // ── Summary ────────────────────────────────────────────────────────────────
   console.log("");
   if (fix && fixesApplied > 0) {
@@ -3266,11 +3286,15 @@ function help() {
     ["npx forgedock disable [dir]", "Opt a directory out of ForgeDock"],
     ["npx forgedock status [dir]", "Show ForgeDock state for a directory"],
     ["npx forgedock run <cmd> [args]", "Run a command headlessly (local claude CLI or Anthropic API)"],
+    ["npx forgedock bootstrap <idea|docs> [--dry-run] [--spike <path>] | --plan <plan.json> [--dry-run]", "Headless /bootstrap: idea to orchestrate-ready backlog (--dry-run previews milestones, titles, DAG and estimated agent cost without GitHub writes); --plan validates and creates (or previews) a plan, and re-running resumes from GitHub"],
     ["npx forgedock run-issue <issue> [--lane <lane>] [--lane-override] [--path core|full] [--retry] [--wait-lease]", "Run the lean core work-on on one issue (default; --path full = legacy multi-phase pipeline; the lane comes from the issue: milestone → milestone/<slug>, else staging; --lane is an assertion, never the default branch) (--retry reopens an engine-error run, or a needs-human run once the label is cleared; --wait-lease waits for a held lease; exit 75 = deferred, lease held)"],
     ["npx forgedock orchestrate <issue...> [--parallel K] [--lane L]", "Run core work-on on several issues with bounded concurrency and Depends-on ordering; each issue on its own lane, cross-lane dependencies reported"],
-    ["npx forgedock orchestrate --resume <id> | --set-parallel <id> K", "Continue a batch after the dispatcher died (live lanes re-adopted), or change its concurrency live"],
+    ["npx forgedock orchestrate --resume <id> | --set-parallel <id> K", "Reattach to a batch hosted by the schedd daemon (a finished one replays its report, an orphaned one is resumed with live lanes re-adopted), or change its concurrency live"],
     ["npx forgedock triage <issue...> [--lane L] [--dry-run]", "Already-fixed pre-filter: close issues a merged PR or green CI already resolved; print the rest"],
-    ["npx forgedock gc", "Prune stale per-run scratch dirs under ~/.forge/tmp/runs (older than 6h, owner dead)"],
+    ["npx forgedock limits show | clear <id>|--all", "Show or clear per-account usage-limit holds shared by orchestrate batches"],
+    ["npx forgedock sched show | cap <N>|--clear | weight <owner/repo> <N>|--clear", "Show the host lane registry, set the host-wide lane cap shared by orchestrate batches, or set a repo's share of slots within a priority (default equal)"],
+    ["npx forgedock schedd [status]", "Host scheduler daemon hosting every orchestrate batch (auto-spawned by orchestrate; run in the foreground for a systemd-user unit); status shows its pid and batches"],
+    ["npx forgedock gc [--apply] [--older-than 6h] [--docker] [--worktrees] [--branches] [--tmp] [--processes] [--sandboxes]", "Report (dry-run default) and reclaim ForgeDock-owned leftovers: stale lane ledgers, forgedock-labelled containers/volumes, lean-*/agent-* worktrees, abandoned lane branches, ~/.forge/tmp scratch, orphaned lane processes, orphaned lane sandboxes; unowned resources and unpushed work are only reported"],
     ["npx forgedock resume-stalled [--dry-run]", "Fleet stall recovery — re-dispatch expired-lease issues"],
     ["npx forgedock demo", "Set up a risk-free demo repo and print next steps"],
     ["npx forgedock labels [setup] [--repo owner/repo] [--domain a,b,c]", "Bootstrap ForgeDock-managed labels on a GitHub repo, plus optional per-repo domain labels (idempotent)"],
@@ -3279,7 +3303,7 @@ function help() {
     ["npx forgedock report [--days 30] [--md] [--json]", "30-day pipeline impact receipts for your repo"],
     ["npx forgedock stats [--since <ver|date>] [--repo owner/repo] [--json] [--refresh]", "Version-over-version cost/time/flow/review/gate report from GitHub telemetry (no model)"],
     ["npx forgedock doctor", "Check installation health"],
-    ["npx forgedock doctor --fix", "Auto-fix deterministic issues (symlinks, hook, labels, legacy block)"],
+    ["npx forgedock doctor --fix", "Auto-fix deterministic issues (symlinks, hook, labels, legacy block) and reclaim ForgeDock-owned leftovers (forge.yaml hygiene TTLs)"],
     ["npx forgedock update", "Pull latest & reinstall"],
     ["npx forgedock uninstall", "Remove commands"],
     ["npx forgedock opencode install", "Install token-efficient native OpenCode commands"],
@@ -3844,10 +3868,10 @@ async function watch() {
 // The journey-routed commands render their own branded marks (hero/compact);
 // splash() would double-brand them. Engine-surface commands, help, and
 // unknown-command output keep the logo.
-const SPLASH_COMMANDS = new Set(["run", "run-issue", "orchestrate", "triage", "gc", "resume-stalled", "demo", "doctor", "watch", "help", "--help", "-h"]);
+const SPLASH_COMMANDS = new Set(["run", "bootstrap", "run-issue", "orchestrate", "triage", "gc", "resume-stalled", "demo", "doctor", "watch", "help", "--help", "-h"]);
 const KNOWN_COMMANDS = new Set([
   "install", "init", "enable", "disable", "status", "uninstall", "update",
-  "run", "backend-check", "run-issue", "orchestrate", "triage", "gc", "resume-stalled", "demo", "doctor", "watch", "labels", "config", "stats", "help", "--help", "-h",
+  "run", "bootstrap", "backend-check", "run-issue", "orchestrate", "triage", "gc", "resume-stalled", "demo", "doctor", "watch", "labels", "config", "stats", "limits", "sched", "schedd", "help", "--help", "-h",
   "version", "--version", "-v",
 ]);
 if (SPLASH_COMMANDS.has(command) || !KNOWN_COMMANDS.has(command)) splash(command);
@@ -4004,6 +4028,18 @@ switch (command) {
   case "run":
     await run();
     break;
+  case "bootstrap": {
+    // #4177: headless /bootstrap; --plan runs scripts/bootstrap-create.mjs directly (no LLM), a re-run resumes.
+    const { runBootstrap } = await import("./bootstrap.mjs");
+    try {
+      const { runCommand } = await import("./runner.mjs");
+      exitCode = await runBootstrap(restArgs, { runCommand, commandsDir: COMMANDS_DIR });
+    } catch (err) {
+      process.stderr.write(`${RED}${err.message}${RESET}\n`);
+      exitCode = 1;
+    }
+    break;
+  }
   case "backend-check": {
     // Exit codes: 0 = ready, 1 = no usable backend, 2 = bad usage (unknown option).
     // Only --quiet is accepted; anything else is a usage error so callers can tell a
@@ -4046,10 +4082,9 @@ switch (command) {
     break;
   }
   case "gc": {
-    // Best-effort prune of core-run scratch dirs (older than 6h, owner pid dead). Never touches dirs we did not create.
-    const { makeRunTmp } = await import("./lean/effects.mjs");
-    const removed = await makeRunTmp().prune();
-    process.stdout.write(`gc: removed ${removed.length} stale run tmp dir(s)\n`);
+    // Reaper (#4180): dry-run report by default; --apply reclaims only ForgeDock-owned resources (ledger, label, lane naming, own tmp).
+    const { runGc } = await import("./gc.mjs");
+    exitCode = await runGc(restArgs);
     break;
   }
   case "orchestrate": {
@@ -4063,6 +4098,37 @@ switch (command) {
     }
     break;
   }
+  case "limits": {
+    const { limitsFromCli } = await import("./engine-cli.mjs");
+    try {
+      await limitsFromCli(restArgs);
+    } catch (err) {
+      process.stderr.write(`${RED}${err.message}${RESET}\n`);
+      exitCode = 1;
+    }
+    break;
+  }
+  case "schedd": {
+    // #4192: the host scheduler daemon; runs until idle or SIGTERM. Lanes are detached and survive it.
+    const { scheddFromCli } = await import("./engine-cli.mjs");
+    try {
+      await scheddFromCli(restArgs);
+    } catch (err) {
+      process.stderr.write(`${RED}${err.message}${RESET}\n`);
+      exitCode = 1;
+    }
+    break;
+  }
+  case "sched": {
+    const { schedFromCli } = await import("./engine-cli.mjs");
+    try {
+      await schedFromCli(restArgs);
+    } catch (err) {
+      process.stderr.write(`${RED}${err.message}${RESET}\n`);
+      exitCode = 1;
+    }
+    break;
+  }
   case "triage": {
     const { triageFromCli } = await import("./engine-cli.mjs");
     try {