opencode-usage-coach 0.7.0 → 0.7.1

package/dist/index.js

@@ -114,6 +114,8 @@ function saveInvestigationResult(keywords, result, source) {
 // src/index.ts
 var PLUGIN_NAME = "opencode-usage-coach";
 var TTL_MS = Number(process.env.UC_TTL_MS ?? 6e4);
+var DEFAULT_MAX_STEPS = Number(process.env.UC_MAX_STEPS ?? 30) || 30;
+var WATCHDOG_POLL_MS = Math.max(1e3, Number(process.env.UC_WATCHDOG_POLL_MS ?? 3e3) || 3e3);
 var PIPE_LOG = join2(homedir(), ".cache", "opencode-usage-coach", "pipeline.log");
 function pipeLog(msg) {
   try {
@@ -205,6 +207,40 @@ function writeHarness(sessionID, h) {
   } catch {
   }
 }
+function updateSubSession(sessionID, taskId, fields) {
+  try {
+    const h = readHarness(sessionID);
+    if (!h) return;
+    const t = h.tasks.find((x) => x.id === taskId);
+    if (!t) return;
+    Object.assign(t, fields);
+    writeHarness(sessionID, h);
+  } catch {
+  }
+}
+function clearSubSession(sessionID, taskId) {
+  try {
+    const h = readHarness(sessionID);
+    if (!h) return;
+    const t = h.tasks.find((x) => x.id === taskId);
+    if (!t) return;
+    t.subSessionId = void 0;
+    t.subStep = void 0;
+    t.lastActivity = void 0;
+    t.subElapsed = void 0;
+    writeHarness(sessionID, h);
+  } catch {
+  }
+}
+function findActiveTaskId(sessionID, status) {
+  try {
+    const h = readHarness(sessionID);
+    if (!h) return void 0;
+    return h.tasks.find((x) => x.status === status)?.id;
+  } catch {
+    return void 0;
+  }
+}
 function readHarnessCfg(dir) {
   const tryRead = (p) => {
     try {
@@ -218,8 +254,16 @@ function readHarnessCfg(dir) {
     ...tryRead(join2(dir, "harness.config.json"))
   };
 }
-async function runModel(client, model, prompt, directory) {
+async function runModel(client, model, prompt, directory, track, maxSteps = DEFAULT_MAX_STEPS) {
   const t0 = Date.now();
+  const subStart = Date.now();
+  let poller = null;
+  let subId = null;
+  let timedOut = false;
+  let signalTimeout;
+  const timeoutSignal = new Promise((resolve2) => {
+    signalTimeout = resolve2;
+  });
   try {
     const slash = model.indexOf("/");
     const providerID = slash >= 0 ? model.slice(0, slash) : model;
@@ -227,12 +271,75 @@ async function runModel(client, model, prompt, directory) {
     const s = await client.session.create({ body: { title: "uc-harness-sub" }, query: { directory } });
     const id = s?.data?.info?.id ?? s?.data?.id ?? s?.id;
     if (!id) return `ERROR: session.create returned no id (response: ${JSON.stringify(s?.data ?? s).slice(0, 200)})`;
-    log(`runModel(${model}): session ${id} created, sending prompt (${prompt.length} chars)`);
-    const resp = await client.session.prompt({
+    subId = id;
+    log(`runModel(${model}): session ${id} created, sending prompt (${prompt.length} chars), max_steps=${maxSteps}`);
+    poller = setInterval(async () => {
+      if (timedOut) return;
+      try {
+        let step = 0;
+        let lastTs = (/* @__PURE__ */ new Date()).toISOString();
+        try {
+          const msgs = await client.session.messages?.({ path: { id } });
+          const msgList = Array.isArray(msgs?.data) ? msgs.data : Array.isArray(msgs) ? msgs : [];
+          if (msgList.length) {
+            step = msgList.filter((m) => {
+              const role = m?.role ?? m?.info?.role;
+              return role === "assistant";
+            }).length;
+            const last = msgList[msgList.length - 1];
+            const ts = last?.ts ?? last?.info?.updatedAt ?? last?.info?.completedAt ?? last?.updatedAt;
+            if (ts) lastTs = String(ts);
+          }
+        } catch {
+        }
+        if (step > maxSteps) {
+          log(`runModel(${model}): STEP LIMIT exceeded (${step} > ${maxSteps}), aborting session ${id}`);
+          timedOut = true;
+          try {
+            await client.session.abort?.({ path: { id } });
+          } catch {
+          }
+          signalTimeout();
+          return;
+        }
+        if (track) {
+          const elapsed2 = Math.round((Date.now() - subStart) / 1e3);
+          updateSubSession(track.sessionID, track.taskId, {
+            subSessionId: id,
+            subStep: step,
+            lastActivity: lastTs,
+            subElapsed: elapsed2
+          });
+        }
+      } catch (e) {
+        log(`runModel poller err: ${String(e)}`);
+      }
+    }, WATCHDOG_POLL_MS);
+    const promptP = client.session.prompt({
       path: { id },
       body: { model: { providerID, modelID }, parts: [{ type: "text", text: prompt }] }
-    });
+    }).then(
+      (r) => r,
+      () => null
+      // abort causes rejection -> return null (handled via timedOut flag)
+    );
+    const resp = await Promise.race([promptP, timeoutSignal.then(() => null)]);
     const elapsed = Math.round((Date.now() - t0) / 1e3);
+    if (timedOut) {
+      try {
+        const summary = await client.session.summarize?.({ path: { id } });
+        log(`runModel(${model}): TIMED OUT summary: ${JSON.stringify(summary?.data ?? summary).slice(0, 300)}`);
+      } catch {
+      }
+      try {
+        await client.session.delete?.({ path: { id } });
+      } catch {
+      }
+      subId = null;
+      log(`runModel(${model}): TIMED OUT after ${elapsed}s (${maxSteps} steps exceeded)`);
+      return `Task appears too large (exceeded ${maxSteps} steps). Consider splitting into smaller subtasks.
+[usage-coach NEXT] split the original task into smaller subtasks (each should complete within ${maxSteps} steps), then re-run generate for each subtask.`;
+    }
     const parts = resp?.data?.parts ?? resp?.parts ?? [];
     const text = parts.filter((p) => p?.type === "text").map((p) => p?.text ?? "").join("");
     try {
@@ -244,12 +351,27 @@ async function runModel(client, model, prompt, directory) {
       await client.session.delete?.({ path: { id } });
     } catch {
     }
+    subId = null;
     log(`runModel(${model}): done ${elapsed}s, ${text.length} chars`);
     return text.trim() || `ERROR: no assistant text in prompt response after ${elapsed}s (parts: ${parts.length}, types: ${parts.map((p) => p?.type).join(",")})`;
   } catch (e) {
     const elapsed = Math.round((Date.now() - t0) / 1e3);
     log(`runModel err (${model}, ${elapsed}s): ${String(e)}`);
     return `ERROR: runModel exception after ${elapsed}s: ${String(e)}`;
+  } finally {
+    if (poller) clearInterval(poller);
+    if (track) {
+      try {
+        clearSubSession(track.sessionID, track.taskId);
+      } catch {
+      }
+    }
+    if (subId) {
+      try {
+        await client.session.delete?.({ path: { id: subId } });
+      } catch {
+      }
+    }
   }
 }
 var HARNESS_AGENTS = (process.env.UC_HARNESS_AGENT ?? "Usage-Coach-Harness").split(",").map((s) => s.trim().toLowerCase()).filter(Boolean);
@@ -549,12 +671,14 @@ async function UsageCoachPlugin(input) {
       // Custom tools for the harness agent mode — report status to the panel.
       tool: {
         harness_start: tool({
-          description: "Start the harness: register the total task count on the panel. Call once when the harness loop begins.",
+          description: "Start the harness: register the total task count on the panel. Call once when the harness loop begins. IMPORTANT: each generate/generate_batch sub-session is step-limited (default 30). If any task seems too large, split it into smaller subtasks BEFORE starting \u2014 oversized tasks will timeout.",
           args: { name: tool.schema.string(), total: tool.schema.number() },
           async execute(args, ctx) {
             writeHarness(ctx.sessionID, { name: args.name, total: args.total, current: 0, tasks: [], usage: {}, active: true, startedAt: (/* @__PURE__ */ new Date()).toISOString() });
             return `Harness '${args.name}' started (${args.total} tasks).
 
+STEP LIMIT (default ${DEFAULT_MAX_STEPS}): each generate call creates a sub-session that is automatically aborted if it exceeds ${DEFAULT_MAX_STEPS} assistant steps. Before starting the loop, review each task: can it be completed in a focused, single-pass effort? If a task seems too broad (multiple files, multiple features, open-ended research), SPLIT it now into 2-3 smaller subtasks. A timeout wastes quota \u2014 split upfront.
+
 DETERMINISTIC LOOP \u2014 first classify the tasks:
   INDEPENDENT = task B does NOT need task A's output  ->  use PATH A (parallel, faster)
   DEPENDENT   = task B needs task A's output          ->  use PATH B (sequential)
@@ -669,7 +793,14 @@ Output a structured root cause:
 category: (one of: constraint-violation, missing-context, tool-misuse, model-limitation, other)
 explanation: <why it failed>
 evidence: <file/line or specific quote>`;
-            const out = await runModel(input.client, cfg.generator, domainPrefix + rcaPrompt, ctx.directory);
+            const invTaskId = findActiveTaskId(ctx.sessionID, "revising");
+            const out = await runModel(
+              input.client,
+              cfg.generator,
+              domainPrefix + rcaPrompt,
+              ctx.directory,
+              invTaskId ? { sessionID: ctx.sessionID, taskId: invTaskId } : void 0
+            );
             if (domainEmpty && keywords.length) {
               try {
                 saveInvestigationResult(keywords, out, "investigate");
@@ -697,7 +828,14 @@ Grade feedback: ${args.gradeResult}
 Diagnosis: ${args.diagnosis}
 Is the diagnosis CORRECT and ACTIONABLE (leads to a useful rule)?
 Output PASS (the diagnosis is right) or FAIL (re-investigate needed), then reason.`;
-            const out = await runModel(input.client, model, verifyPrompt, ctx.directory);
+            const verTaskId = findActiveTaskId(ctx.sessionID, "revising");
+            const out = await runModel(
+              input.client,
+              model,
+              verifyPrompt,
+              ctx.directory,
+              verTaskId ? { sessionID: ctx.sessionID, taskId: verTaskId } : void 0
+            );
             let verdict = "FAIL";
             if (!out.startsWith("ERROR:")) {
               const f = (out.split("\n").find((l) => l.trim()) ?? "").trim();
@@ -724,7 +862,14 @@ Diagnosis: ${args.diagnosis}
 Failed task: ${args.task}
 Output a single rule in the form: 'For <task-type> tasks, always <check/do X> because <reason>.'
 Keep it concrete and actionable.`;
-            const out = await runModel(input.client, cfg.generator, genPrompt, ctx.directory);
+            const genRuleTaskId = findActiveTaskId(ctx.sessionID, "revising");
+            const out = await runModel(
+              input.client,
+              cfg.generator,
+              genPrompt,
+              ctx.directory,
+              genRuleTaskId ? { sessionID: ctx.sessionID, taskId: genRuleTaskId } : void 0
+            );
             const rule = out;
             try {
               const date = (/* @__PURE__ */ new Date()).toISOString().slice(0, 10);
@@ -744,8 +889,8 @@ Origin: ${args.task}
         // Per-role model execution (config-driven, quota-aware, same server, no deadlock).
         // P1: quota decision drives model selection + concurrency.
         generate: tool({
-          description: "Run the GENERATOR model on a prompt. Quota-aware: on THROTTLE, auto-switches to lighterModel if configured. Returns the model's text response.",
-          args: { prompt: tool.schema.string() },
+          description: "Run the GENERATOR model on a prompt. Quota-aware: on THROTTLE, auto-switches to lighterModel if configured. Returns the model's text response. Step-limited: aborts after max_steps (default 30) to prevent runaway tasks.",
+          args: { prompt: tool.schema.string(), max_steps: tool.schema.number().optional().describe("Maximum sub-session steps before timeout (default 30). Increase for complex tasks, decrease to fail fast on scope creep.") },
           async execute(args, ctx) {
             const cfg = readHarnessCfg(ctx.directory);
             if (!cfg.generator) return 'ERROR: no generator model configured. Set "generator" in harness.config.json (see harness.config.example.json).';
@@ -781,22 +926,33 @@ ${rules}
             } catch (e) {
               log(`generate domain query err: ${String(e)}`);
             }
-            const out = await runModel(input.client, model, prefix + args.prompt, ctx.directory);
-            if (domainEmpty && keywords.length) {
+            const genTaskId = findActiveTaskId(ctx.sessionID, "generating");
+            const maxSteps = args.max_steps ?? DEFAULT_MAX_STEPS;
+            const out = await runModel(
+              input.client,
+              model,
+              prefix + args.prompt,
+              ctx.directory,
+              genTaskId ? { sessionID: ctx.sessionID, taskId: genTaskId } : void 0,
+              maxSteps
+            );
+            const isTimeoutOrError = out.startsWith("Task appears too large") || out.startsWith("ERROR:");
+            if (domainEmpty && keywords.length && !isTimeoutOrError) {
               try {
                 saveInvestigationResult(keywords, out, "generate");
               } catch (e) {
                 log(`generate save err: ${String(e)}`);
               }
             }
+            if (out.startsWith("Task appears too large")) return out;
             return out + (throttle ? `
 [usage-coach] quota THROTTLE \u2014 used lighter model ${cfg.lighterModel}` : "") + `
 [usage-coach NEXT] call task_update(i, title, "grading"), then grade to evaluate this work.`;
           }
         }),
         generate_batch: tool({
-          description: "Run the GENERATOR model on MULTIPLE tasks. Quota-aware: GO = full parallel; THROTTLE = lighter model + concurrency capped at 2; STOP = refused. Use for INDEPENDENT tasks.",
-          args: { tasks: tool.schema.array(tool.schema.object({ id: tool.schema.number(), prompt: tool.schema.string() })) },
+          description: "Run the GENERATOR model on MULTIPLE tasks. Quota-aware: GO = full parallel; THROTTLE = lighter model + concurrency capped at 2; STOP = refused. Use for INDEPENDENT tasks. Step-limited: each sub-session aborts after max_steps (default 30).",
+          args: { tasks: tool.schema.array(tool.schema.object({ id: tool.schema.number(), prompt: tool.schema.string() })), max_steps: tool.schema.number().optional().describe("Maximum sub-session steps per task before timeout (default 30).") },
           async execute(args, ctx) {
             const cfg = readHarnessCfg(ctx.directory);
             if (!cfg.generator) return 'ERROR: no generator model configured. Set "generator" in harness.config.json (see harness.config.example.json).';
@@ -813,7 +969,14 @@ ${rules}
             for (let i = 0; i < args.tasks.length; i += limit) {
               const batch = args.tasks.slice(i, i + limit);
               const out = await Promise.all(batch.map(async (t) => {
-                const r = await runModel(input.client, model, t.prompt, ctx.directory);
+                const r = await runModel(
+                  input.client,
+                  model,
+                  t.prompt,
+                  ctx.directory,
+                  { sessionID: ctx.sessionID, taskId: t.id },
+                  args.max_steps ?? DEFAULT_MAX_STEPS
+                );
                 return `[task ${t.id}] ${r}`;
               }));
               results.push(...out);
@@ -830,7 +993,14 @@ ${rules}
             const cfg = readHarnessCfg(ctx.directory);
             const model = cfg.grader ?? cfg.generator;
             if (!model) return "FAIL\n(ERROR: no grader/generator model configured.)\n[usage-coach NEXT] configure grader in harness.config.json, then retry grade.";
-            const out = await runModel(input.client, model, args.prompt, ctx.directory);
+            const gradeTaskId = findActiveTaskId(ctx.sessionID, "grading");
+            const out = await runModel(
+              input.client,
+              model,
+              args.prompt,
+              ctx.directory,
+              gradeTaskId ? { sessionID: ctx.sessionID, taskId: gradeTaskId } : void 0
+            );
             let verdict = "FAIL";
             if (!out.startsWith("ERROR:")) {
               const f = (out.split("\n").find((l) => l.trim()) ?? "").trim();

package/dist/tui.js

@@ -391,8 +391,14 @@ function initializeTui(api, disposeRoot) {
         const lbl = TLABEL[t.status] ?? t.status;
         const rev = t.revisions > 0 && t.status === "revising" ? `(${t.revisions})` : "";
         const mdl = t.model ? ` ${t.model.split("/").pop() ?? t.model}` : "";
+        const hasSub = !!t.subSessionId;
+        const subStepStr = hasSub && t.subStep !== void 0 && t.subStep > 0 ? ` step:${t.subStep}` : "";
+        const subEl = hasSub && t.subElapsed !== void 0 ? ` ${t.subElapsed}s` : "";
+        const subWarn = hasSub && (t.subElapsed ?? 0) > 300;
         const elapsed = t.startedAt ? Math.max(0, Math.round((Date.now() - new Date(t.startedAt).getTime()) / 1e3)) : 0;
-        const elapsedStr = t.status === "completed" || t.status === "failed" ? "" : elapsed > 0 ? ` ${elapsed}s` : "";
+        const taskEl = t.status === "completed" || t.status === "failed" ? "" : elapsed > 0 ? ` ${elapsed}s` : "";
+        const displayEl = hasSub ? subEl : taskEl;
+        const lineKey = subWarn ? "warning" : sKey;
         nodes.push((() => {
           var _el$55 = _$createElement("text"), _el$56 = _$createTextNode(` \u25CF `), _el$57 = _$createTextNode(` `), _el$58 = _$createTextNode(` `);
           _$insertNode(_el$55, _el$56);
@@ -402,9 +408,10 @@ function initializeTui(api, disposeRoot) {
           _$insert(_el$55, mdl, _el$57);
           _$insert(_el$55, lbl, _el$58);
           _$insert(_el$55, rev, _el$58);
-          _$insert(_el$55, elapsedStr, _el$58);
+          _$insert(_el$55, subStepStr, _el$58);
+          _$insert(_el$55, displayEl, _el$58);
           _$insert(_el$55, () => t.title, null);
-          _$effect((_$p) => _$setProp(_el$55, "style", st(sKey), _$p));
+          _$effect((_$p) => _$setProp(_el$55, "style", st(lineKey), _$p));
           return _el$55;
         })());
         const pv = t.model ? (t.model.split("/")[0] ?? "").split("-")[0] : "";

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "opencode-usage-coach",
-  "version": "0.7.0",
+  "version": "0.7.1",
   "description": "opencode closed-loop usage coach — quota SENSE -> coaching DECIDE -> loop ACT + TUI integration",
   "type": "module",
   "main": "./dist/index.js",