konoha-mcp 2.2.23 → 2.2.24
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/NOTICE +38 -0
- package/README.md +3 -3
- package/apps/web/build/adapter-node.js +1 -1
- package/apps/web/build/build-meta.json +2 -2
- package/apps/web/build/client/_app/immutable/entry/{app.TYH87iYy.js → app.CKJ0NhrX.js} +2 -2
- package/apps/web/build/client/_app/immutable/entry/app.CKJ0NhrX.js.br +0 -0
- package/apps/web/build/client/_app/immutable/entry/app.CKJ0NhrX.js.gz +0 -0
- package/apps/web/build/client/_app/immutable/nodes/{0.D7dPIEWA.js → 0.7A6fQJ6z.js} +1 -1
- package/apps/web/build/client/_app/immutable/nodes/0.7A6fQJ6z.js.br +0 -0
- package/apps/web/build/client/_app/immutable/nodes/{0.D7dPIEWA.js.gz → 0.7A6fQJ6z.js.gz} +0 -0
- package/apps/web/build/client/_app/immutable/nodes/{13.TOXVaTBr.js → 13.CQB_Ed6W.js} +1 -1
- package/apps/web/build/client/_app/immutable/nodes/13.CQB_Ed6W.js.br +0 -0
- package/apps/web/build/client/_app/immutable/nodes/{13.TOXVaTBr.js.gz → 13.CQB_Ed6W.js.gz} +0 -0
- package/apps/web/build/client/_app/version.json +1 -1
- package/apps/web/build/client/_app/version.json.br +0 -0
- package/apps/web/build/client/_app/version.json.gz +0 -0
- package/apps/web/build/server/chunks/instance.js +3 -3
- package/apps/web/build/server/chunks/internal.js +1 -1
- package/apps/web/build/server/entries/pages/_layout.svelte.js +1 -1
- package/apps/web/build/server/entries/pages/_layout.svelte.js.map +1 -1
- package/apps/web/build/server/entries/pages/docs/_page.svelte.js +1 -1
- package/apps/web/build/server/entries/pages/docs/_page.svelte.js.map +1 -1
- package/apps/web/build/server/manifest-full.js +1 -1
- package/apps/web/build/server/manifest.js +1 -1
- package/apps/web/build/server/nodes/0.js +1 -1
- package/apps/web/build/server/nodes/13.js +1 -1
- package/apps/web/build/server/server.js +1 -1
- package/apps/web/package.json +1 -1
- package/bin/commands/bridge.js +62 -4
- package/bin/commands/doctor.js +2 -0
- package/bin/commands/doctor_services.js +57 -0
- package/bin/commands/init.js +16 -5
- package/bin/commands/init_clients.js +95 -8
- package/bin/lib/config_defaults.js +3 -0
- package/bin/lib/runtime_install.js +141 -54
- package/package.json +2 -1
- package/patches/gpt-tokenizer@4.1.0.patch +15 -0
- package/src/agent_contract.js +1 -1
- package/src/agent_deployer.js +48 -35
- package/src/agent_file_safety.js +57 -0
- package/src/bridge/gateway.js +2 -0
- package/src/bridge/gateway_adapters.js +155 -29
- package/src/bridge/handlers/anthropic.js +95 -26
- package/src/bridge/handlers/openai.js +18 -5
- package/src/bridge/sidecar/raw.js +10 -2
- package/src/bridge/stream_usage.js +66 -0
- package/src/client_commands.js +190 -0
- package/src/db_bridges.js +53 -7
- package/src/db_schema.js +11 -2
- package/src/file_tools_mcp.js +17 -9
- package/src/global_skill_sync.js +3 -1
- package/src/mcp/langgraph/edges/router.js +3 -0
- package/src/mcp/langgraph/index.js +22 -3
- package/src/mcp/langgraph/nodes/dispatch.js +3 -0
- package/src/mcp/langgraph/nodes/execute.js +87 -13
- package/src/mcp/langgraph/nodes/review.js +13 -1
- package/src/mcp/langgraph/slice_executor_registry.js +22 -0
- package/src/mcp/langgraph/state.js +5 -0
- package/src/mcp/lazy_subagent.js +11 -1
- package/src/mcp/memory_project_context.js +4 -1
- package/src/mcp/memory_subagent_runner.js +2 -1
- package/src/mcp/skills_common.js +1 -0
- package/src/mcp/subagent_spawner.js +7 -0
- package/src/mcp/todo_walkthrough_sync.js +7 -0
- package/src/mcp/tool_dispatch.js +16 -0
- package/src/mcp/workflow_completion_helpers.js +6 -1
- package/src/mcp/workflow_gates.js +118 -5
- package/src/mcp/workflow_sannin.js +9 -3
- package/src/mcp_tool_manifest.json +5 -1
- package/src/provider_usage_logger.js +92 -0
- package/src/schema_ddl.js +17 -1
- package/src/templates/AGENTS.md +1 -1
- package/src/templates/GEMINI.md +1 -1
- package/src/templates/skills/kage-skill/SKILL.md +1 -1
- package/src/templates/skills/konoha/SKILL.md +23 -165
- package/src/templates/skills/konoha/references/contract.md +1 -1
- package/src/templates/skills/konoha/references/peragent-usage.md +1 -1
- package/src/token_usage.js +57 -0
- package/src/tools_savings_logger.js +26 -8
- package/src/web/controllers/core_controller.js +10 -1
- package/src/workflow/jonin-builder/live_browser_monitor.js +23 -7
- package/apps/web/build/client/_app/immutable/entry/app.TYH87iYy.js.br +0 -0
- package/apps/web/build/client/_app/immutable/entry/app.TYH87iYy.js.gz +0 -0
- package/apps/web/build/client/_app/immutable/nodes/0.D7dPIEWA.js.br +0 -0
- package/apps/web/build/client/_app/immutable/nodes/13.TOXVaTBr.js.br +0 -0
- package/src/templates/skills/konoha/references/architecture.md +0 -99
- package/src/templates/skills/konoha/references/benchmark.md +0 -123
- package/src/templates/skills/konoha/references/docs.md +0 -46
- package/src/templates/skills/konoha/references/test.md +0 -71
- package/src/templates/skills/konoha/references/workflow.md +0 -104
|
@@ -47,6 +47,10 @@ function normalizeTodoItems(tasks = [], status = {}) {
|
|
|
47
47
|
if (rawStatus === 'completed' || rawStatus === 'done') mappedStatus = 'completed';
|
|
48
48
|
else if (rawStatus === 'in_progress' || rawStatus === 'running') mappedStatus = 'in_progress';
|
|
49
49
|
else if (['unverified', 'failed', 'rejected', 'blocked'].includes(rawStatus)) mappedStatus = rawStatus;
|
|
50
|
+
// `spawn_pending` is a real state, not a typo: the slice has a directive the
|
|
51
|
+
// client must run and is only completed by `report_from_agent`. Mapping it to
|
|
52
|
+
// `pending` here erased the marker on every sync and hid the waiting spawn.
|
|
53
|
+
else if (rawStatus === 'spawn_pending') mappedStatus = 'spawn_pending';
|
|
50
54
|
const text = base.task || base.content || base.description || `Step ${idx + 1}`;
|
|
51
55
|
return {
|
|
52
56
|
...base,
|
|
@@ -196,6 +200,9 @@ function applyAgentReportToTodos(workflowStatus, cleanAgent, dispatchId, finalSt
|
|
|
196
200
|
|
|
197
201
|
if (finalStatus === 'completed' || finalStatus === 'unverified') {
|
|
198
202
|
task.status = finalStatus;
|
|
203
|
+
// A report without passing validation evidence cannot have produced a
|
|
204
|
+
// finished child: `completed` requires the evidence gate to have passed.
|
|
205
|
+
task.spawn_state = finalStatus === 'completed' ? 'completed' : 'directive_only';
|
|
199
206
|
task.result = updateData.summary || '';
|
|
200
207
|
task.validation = updateData.validation || [];
|
|
201
208
|
task.verified = updateData.verified;
|
package/src/mcp/tool_dispatch.js
CHANGED
|
@@ -326,6 +326,22 @@ function _mergeSanninGraphResponse(parsedSannin, graphInitRes, threadId) {
|
|
|
326
326
|
};
|
|
327
327
|
}
|
|
328
328
|
|
|
329
|
+
// A slice that only has a spawn directive is not routed work: the client has to
|
|
330
|
+
// run the directive and report back. Answering it with the routing envelope made
|
|
331
|
+
// the caller believe a subagent had been dispatched.
|
|
332
|
+
if (graphInitRes.spawnPending) {
|
|
333
|
+
return {
|
|
334
|
+
...parsedSannin,
|
|
335
|
+
status: 'spawn_pending',
|
|
336
|
+
phase: graphInitRes.phase,
|
|
337
|
+
spawn_pending: graphInitRes.spawn_pending || (wfStatus && wfStatus.spawn_pending) || [],
|
|
338
|
+
next_action: 'spawn_and_report',
|
|
339
|
+
resume_node: 'node_execute',
|
|
340
|
+
langgraph_checkpoint: checkpoint,
|
|
341
|
+
workflow_state: typeof graphInitRes.status === 'object' ? graphInitRes.status : undefined
|
|
342
|
+
};
|
|
343
|
+
}
|
|
344
|
+
|
|
329
345
|
if (parsedSannin.status === 'routed') {
|
|
330
346
|
return {
|
|
331
347
|
...graphInitRes,
|
|
@@ -150,7 +150,12 @@ function extractModifiedFiles(tasks) {
|
|
|
150
150
|
function formatTaskSummary(tasks) {
|
|
151
151
|
const list = (tasks || []).slice(0, 5);
|
|
152
152
|
return list.map(t => {
|
|
153
|
-
|
|
153
|
+
// Executor slices report `summary`; agent reports use `result`. Either may
|
|
154
|
+
// arrive as a non-string (object payload), which used to throw here and
|
|
155
|
+
// took the whole review node down.
|
|
156
|
+
const raw = typeof t.result === 'string' ? t.result
|
|
157
|
+
: (typeof t.summary === 'string' ? t.summary : 'Completed');
|
|
158
|
+
const res = raw.replace(/\n+/g, ' ').trim();
|
|
154
159
|
const short = res.length > 60 ? res.slice(0, 57) + '...' : res;
|
|
155
160
|
return `• [${t.agent || 'agent'}] ${t.task}: ${short}`;
|
|
156
161
|
}).join('\n');
|
|
@@ -1,7 +1,10 @@
|
|
|
1
1
|
'use strict';
|
|
2
2
|
|
|
3
|
+
const fs = require('fs');
|
|
3
4
|
const path = require('path');
|
|
5
|
+
const crypto = require('crypto');
|
|
4
6
|
const { saveWorkflowStatus, cleanupTransientScratchFiles } = require('./workflow_common');
|
|
7
|
+
const { extractModifiedFiles, formatTaskSummary } = require('./workflow_completion_helpers');
|
|
5
8
|
const { syncTodoAndWalkthrough } = require('./todo_walkthrough_sync');
|
|
6
9
|
const { buildPickerBinding } = require('./hitl_picker');
|
|
7
10
|
const { slicePreservingHeadMiddleTail } = require('../context_budget');
|
|
@@ -129,6 +132,8 @@ function buildPromptPayload({
|
|
|
129
132
|
title,
|
|
130
133
|
question,
|
|
131
134
|
choices,
|
|
135
|
+
implemented,
|
|
136
|
+
modified_files: modifiedFiles,
|
|
132
137
|
options
|
|
133
138
|
}) {
|
|
134
139
|
return {
|
|
@@ -139,6 +144,8 @@ function buildPromptPayload({
|
|
|
139
144
|
input_mode: 'reply_with_option_label',
|
|
140
145
|
choices,
|
|
141
146
|
refinement_option: CUSTOM_CHOICE_LABEL,
|
|
147
|
+
...(implemented === undefined ? {} : { implemented }),
|
|
148
|
+
...(modifiedFiles === undefined ? {} : { modified_files: modifiedFiles }),
|
|
142
149
|
options
|
|
143
150
|
};
|
|
144
151
|
}
|
|
@@ -182,8 +189,95 @@ function _serializeHitlPromptEnvelope(baseFields, { stage, gateName, choices, st
|
|
|
182
189
|
});
|
|
183
190
|
}
|
|
184
191
|
|
|
192
|
+
/**
|
|
193
|
+
* Gate 1 is the only moment the approved plan exists as text before execution.
|
|
194
|
+
* Persist the compiled plan as the authoritative plan.md artifact next to
|
|
195
|
+
* `final_report.md` so it survives compaction and the approver can open it.
|
|
196
|
+
*
|
|
197
|
+
* Task dirs are shared with clients, so `plan.md` may be pre-created by someone
|
|
198
|
+
* else — possibly as a symlink or a hard link to a file outside the task dir.
|
|
199
|
+
* Validation-by-path alone is racy, so two rules apply: an existing artifact is
|
|
200
|
+
* only accepted after the opened descriptor proves it is the same regular,
|
|
201
|
+
* singly linked file that was inspected, and the new text is written to a fresh
|
|
202
|
+
* inode that is then renamed over `plan.md`. A rename replaces the directory
|
|
203
|
+
* entry without following it, so even a link that slipped past inspection can
|
|
204
|
+
* never carry the bytes into another file.
|
|
205
|
+
* Write failures must never block the gate itself.
|
|
206
|
+
*/
|
|
207
|
+
const PLAN_ARTIFACT_NAME = 'plan.md';
|
|
208
|
+
|
|
209
|
+
/** True when `plan.md` is absent or a regular, singly linked file we may replace. */
|
|
210
|
+
function _isReplaceableArtifact(planPath, nofollow) {
|
|
211
|
+
let before;
|
|
212
|
+
try {
|
|
213
|
+
before = fs.lstatSync(planPath);
|
|
214
|
+
} catch (err) {
|
|
215
|
+
if (err.code === 'ENOENT') return true;
|
|
216
|
+
throw err;
|
|
217
|
+
}
|
|
218
|
+
// lstat does not follow the final component, so a symlink — dangling or not —
|
|
219
|
+
// and a directory or device all fail here instead of being written through.
|
|
220
|
+
if (!before.isFile()) return false;
|
|
221
|
+
const fd = fs.openSync(planPath, fs.constants.O_WRONLY | nofollow);
|
|
222
|
+
try {
|
|
223
|
+
const opened = fs.fstatSync(fd);
|
|
224
|
+
// Opening by path can land on a different inode than lstat inspected (a
|
|
225
|
+
// swap in between), so the descriptor has to prove its identity. O_NOFOLLOW
|
|
226
|
+
// makes the open itself refuse a link where the platform supports it; the
|
|
227
|
+
// inode comparison covers platforms that do not.
|
|
228
|
+
if (!opened.isFile() || opened.dev !== before.dev || opened.ino !== before.ino) return false;
|
|
229
|
+
// A reported inode of 0 is not an identity, and without O_NOFOLLOW nothing
|
|
230
|
+
// else proves the descriptor is the inspected file.
|
|
231
|
+
if (!nofollow && (before.ino === 0 || opened.ino === 0)) return false;
|
|
232
|
+
// A hard link shares the inode, so every check above still passes while a
|
|
233
|
+
// truncate would land in a file outside the task dir. The artifact is only
|
|
234
|
+
// ever created singly linked, so a second name belongs to someone else.
|
|
235
|
+
// Refusing is the conservative choice; the rename below would be safe even
|
|
236
|
+
// where a filesystem under-reports this count.
|
|
237
|
+
return opened.nlink === 1;
|
|
238
|
+
} finally {
|
|
239
|
+
fs.closeSync(fd);
|
|
240
|
+
}
|
|
241
|
+
}
|
|
242
|
+
|
|
243
|
+
/** Write `text` to a fresh inode and rename it over `planPath`. */
|
|
244
|
+
function _replacePlanArtifact(planPath, text) {
|
|
245
|
+
const tmpPath = `${planPath}.${process.pid}.${crypto.randomUUID()}.tmp`;
|
|
246
|
+
const fd = fs.openSync(tmpPath, fs.constants.O_WRONLY | fs.constants.O_CREAT | fs.constants.O_EXCL, 0o644);
|
|
247
|
+
try {
|
|
248
|
+
fs.writeFileSync(fd, text, 'utf8');
|
|
249
|
+
} finally {
|
|
250
|
+
fs.closeSync(fd);
|
|
251
|
+
}
|
|
252
|
+
try {
|
|
253
|
+
fs.renameSync(tmpPath, planPath);
|
|
254
|
+
} catch (err) {
|
|
255
|
+
try { fs.unlinkSync(tmpPath); } catch (_) { /* temp file already gone */ }
|
|
256
|
+
throw err;
|
|
257
|
+
}
|
|
258
|
+
}
|
|
259
|
+
|
|
260
|
+
function writePlanArtifact(resolvedTaskDir, planText) {
|
|
261
|
+
const text = String(planText === undefined || planText === null ? '' : planText);
|
|
262
|
+
if (!resolvedTaskDir || !text.trim()) return null;
|
|
263
|
+
const nofollow = fs.constants.O_NOFOLLOW || 0;
|
|
264
|
+
try {
|
|
265
|
+
fs.mkdirSync(resolvedTaskDir, { recursive: true });
|
|
266
|
+
// Canonicalize after creation so a directory symlink resolves to the real
|
|
267
|
+
// task dir and the artifact can never be placed relative to a link.
|
|
268
|
+
const taskDir = fs.realpathSync(resolvedTaskDir);
|
|
269
|
+
const planPath = path.join(taskDir, PLAN_ARTIFACT_NAME);
|
|
270
|
+
if (!_isReplaceableArtifact(planPath, nofollow)) return null;
|
|
271
|
+
_replacePlanArtifact(planPath, text);
|
|
272
|
+
return planPath;
|
|
273
|
+
} catch (_) {
|
|
274
|
+
return null;
|
|
275
|
+
}
|
|
276
|
+
}
|
|
277
|
+
|
|
185
278
|
function buildGate1Prompt(status, resolvedTaskDir, rootTaskId, planContent, options = {}) {
|
|
186
279
|
const planText = planContent || status.plan || '';
|
|
280
|
+
const planPath = writePlanArtifact(resolvedTaskDir, planText);
|
|
187
281
|
const choices = [...GATE1_CHOICES, CUSTOM_CHOICE_LABEL];
|
|
188
282
|
const promptPayload = buildPromptPayload({
|
|
189
283
|
gate: 'gate_1',
|
|
@@ -214,14 +308,15 @@ function buildGate1Prompt(status, resolvedTaskDir, rootTaskId, planContent, opti
|
|
|
214
308
|
]
|
|
215
309
|
});
|
|
216
310
|
|
|
217
|
-
const message = 'Action plan compiled. Execution paused awaiting human plan approval (Gate 1).\n
|
|
218
|
-
|
|
311
|
+
const message = 'Action plan compiled. Execution paused awaiting human plan approval (Gate 1).\n' +
|
|
312
|
+
(planPath ? `\nFull plan written to ${planPath}\n\n${String(planText).trim()}\n` : '\n' + String(planText).trim() + '\n') +
|
|
313
|
+
'\nReply with the option label (or its number):\n' +
|
|
219
314
|
renderChoiceList(choices) +
|
|
220
315
|
'\n\nChoosing "' + CUSTOM_CHOICE_LABEL + '" lets you type any instruction, e.g.: ' +
|
|
221
316
|
CUSTOM_CHOICE_LABEL + ': use Postgres instead of SQLite';
|
|
222
317
|
|
|
223
318
|
return _serializeHitlPromptEnvelope(
|
|
224
|
-
{ status: 'pending_plan_approval', phase: 'plan_approval', plan: planText, tasks: status.tasks || [] },
|
|
319
|
+
{ status: 'pending_plan_approval', phase: 'plan_approval', plan: planText, plan_path: planPath, tasks: status.tasks || [] },
|
|
225
320
|
{ stage: 'gate1', gateName: 'Gate 1 (Plan Approval)', choices, status, resolvedTaskDir, taskId: rootTaskId, message, promptPayload, client: options.client }
|
|
226
321
|
);
|
|
227
322
|
}
|
|
@@ -230,6 +325,16 @@ function buildGate2Prompt(status, resolvedTaskDir, rootId, options = {}) {
|
|
|
230
325
|
const confidenceVal = options.confidence !== undefined ? options.confidence : 100;
|
|
231
326
|
const reportPath = options.report;
|
|
232
327
|
const finalReport = reportPath || status?.report || '';
|
|
328
|
+
// Gate 2 asks a human to accept shipped work. Enumerate what actually landed
|
|
329
|
+
// so the approval is about the delivery, not only the confidence number.
|
|
330
|
+
const tasks = Array.isArray(status && status.tasks) ? status.tasks : [];
|
|
331
|
+
const doneTasks = tasks.filter(t => {
|
|
332
|
+
const s = String((t && t.status) || '').toLowerCase();
|
|
333
|
+
return s === 'completed' || s === 'done';
|
|
334
|
+
});
|
|
335
|
+
const implemented = formatTaskSummary(doneTasks) || 'No completed tasks recorded.';
|
|
336
|
+
const modifiedFiles = extractModifiedFiles(doneTasks);
|
|
337
|
+
const fileList = modifiedFiles.length > 0 ? modifiedFiles.map(f => `- ${f}`).join('\n') : '- none reported';
|
|
233
338
|
const choices = [...GATE2_CHOICES, CUSTOM_CHOICE_LABEL];
|
|
234
339
|
const promptPayload = buildPromptPayload({
|
|
235
340
|
gate: 'gate_2',
|
|
@@ -237,6 +342,8 @@ function buildGate2Prompt(status, resolvedTaskDir, rootId, options = {}) {
|
|
|
237
342
|
title: 'Gate 2: Delivery Sign-Off Required',
|
|
238
343
|
question: `Kage review completed (${confidenceVal}% Confidence). Delivery paused awaiting human sign-off (Gate 2). How would you like to proceed?`,
|
|
239
344
|
choices,
|
|
345
|
+
implemented,
|
|
346
|
+
modified_files: modifiedFiles,
|
|
240
347
|
options: [
|
|
241
348
|
{
|
|
242
349
|
id: '1',
|
|
@@ -261,13 +368,15 @@ function buildGate2Prompt(status, resolvedTaskDir, rootId, options = {}) {
|
|
|
261
368
|
});
|
|
262
369
|
|
|
263
370
|
const message = `Kage review completed (${confidenceVal}% Confidence). Delivery paused awaiting human sign-off (Gate 2).\n\n` +
|
|
371
|
+
`Implemented:\n${implemented}\n\n` +
|
|
372
|
+
`Modified files:\n${fileList}\n\n` +
|
|
264
373
|
'Reply with the option label (or its number):\n' +
|
|
265
374
|
renderChoiceList(choices) +
|
|
266
375
|
'\n\nChoosing "' + CUSTOM_CHOICE_LABEL + '" lets you type the exact fix you want, e.g.: ' +
|
|
267
376
|
CUSTOM_CHOICE_LABEL + ': add integration tests for the upload path';
|
|
268
377
|
|
|
269
378
|
return _serializeHitlPromptEnvelope(
|
|
270
|
-
{ status: 'pending_human_signoff', phase: 'delivery_signoff', confidence: confidenceVal, report: finalReport },
|
|
379
|
+
{ status: 'pending_human_signoff', phase: 'delivery_signoff', confidence: confidenceVal, report: finalReport, implemented, modified_files: modifiedFiles },
|
|
271
380
|
{ stage: 'gate2', gateName: 'Gate 2 (Delivery Sign-Off)', choices, status, resolvedTaskDir, taskId: rootId, message, promptPayload, client: options.client }
|
|
272
381
|
);
|
|
273
382
|
}
|
|
@@ -493,5 +602,9 @@ module.exports = {
|
|
|
493
602
|
parseGateChoice,
|
|
494
603
|
applyGate1Choice,
|
|
495
604
|
applyGate2Choice,
|
|
496
|
-
compactPostApprovalContext
|
|
605
|
+
compactPostApprovalContext,
|
|
606
|
+
// Internals, exported so the platform-specific branches can be exercised on a
|
|
607
|
+
// host that has O_NOFOLLOW (Windows coverage without a Windows runner).
|
|
608
|
+
_isReplaceableArtifact,
|
|
609
|
+
_replacePlanArtifact
|
|
497
610
|
};
|
|
@@ -593,11 +593,17 @@ function _finalizeSanninDispatch(prep, envelope) {
|
|
|
593
593
|
: '';
|
|
594
594
|
|
|
595
595
|
const currentSt = loadWorkflowStatus(resolvedTaskDir);
|
|
596
|
+
// Tasks are owned by the plan (Gate 1) and by the slices that report into them.
|
|
597
|
+
// Re-seeding them from the route on every call reset finished work to pending, so
|
|
598
|
+
// a resumed call re-dispatched a slice whose child had already run.
|
|
599
|
+
const existingTasks = Array.isArray(currentSt.tasks) ? currentSt.tasks : [];
|
|
596
600
|
const syncBundle = syncTodoAndWalkthrough(currentSt, {
|
|
597
601
|
client: actClient, taskDir: resolvedTaskDir, taskId,
|
|
598
|
-
tasks:
|
|
599
|
-
?
|
|
600
|
-
:
|
|
602
|
+
tasks: existingTasks.length > 0
|
|
603
|
+
? undefined
|
|
604
|
+
: (parallelWorkstreams.length > 1
|
|
605
|
+
? parallelWorkstreams.map((ws, i) => ({ id: `task_${i + 1}`, task: `${effPrompt} (${ws.domain} slice)`, agent: ws.agent, status: 'pending' }))
|
|
606
|
+
: [{ id: 'task_1', task: effPrompt, agent: selectedAgent, status: 'in_progress' }])
|
|
601
607
|
});
|
|
602
608
|
const advisorSupervisor = initAdvisorSessionSupervisor({ client: spawnClient, status: currentSt, goal: effPrompt, todos: syncBundle.todos });
|
|
603
609
|
saveWorkflowStatus(resolvedTaskDir, currentSt);
|
|
@@ -1043,7 +1043,8 @@
|
|
|
1043
1043
|
],
|
|
1044
1044
|
"description": "Compute device preference; falls back to CPU when no GPU is available."
|
|
1045
1045
|
}
|
|
1046
|
-
}
|
|
1046
|
+
},
|
|
1047
|
+
"additionalProperties": false
|
|
1047
1048
|
}
|
|
1048
1049
|
},
|
|
1049
1050
|
{
|
|
@@ -1290,6 +1291,9 @@
|
|
|
1290
1291
|
"dispatch_id": {
|
|
1291
1292
|
"type": "string"
|
|
1292
1293
|
},
|
|
1294
|
+
"task_id": {
|
|
1295
|
+
"type": "string"
|
|
1296
|
+
},
|
|
1293
1297
|
"validation": {
|
|
1294
1298
|
"type": "array",
|
|
1295
1299
|
"items": {
|
|
@@ -0,0 +1,92 @@
|
|
|
1
|
+
'use strict';
|
|
2
|
+
|
|
3
|
+
/**
|
|
4
|
+
* src/provider_usage_logger.js — persist provider-reported token usage.
|
|
5
|
+
*
|
|
6
|
+
* `tool_calls` measures how much a Konoha tool saved in bytes; it is not a
|
|
7
|
+
* record of what an LLM provider billed. Gateway traffic (`/v1/messages` ->
|
|
8
|
+
* bridge `/v1/chat/completions`) never produced a tool row at all, so
|
|
9
|
+
* `tool_calls.input_tokens` was NULL for every real runtime write and the
|
|
10
|
+
* budget report could only ever print estimates.
|
|
11
|
+
*
|
|
12
|
+
* This writer fills that hole without touching tool-call accounting:
|
|
13
|
+
* `queryStats`/`getSavingsReport` count every `tool_calls` row as a call, so a
|
|
14
|
+
* pseudo-row per gateway request would have inflated savings and per-client
|
|
15
|
+
* attribution. Rows land in `provider_usage` instead, which no savings query
|
|
16
|
+
* reads.
|
|
17
|
+
*
|
|
18
|
+
* Only a provider payload is ever written. When the provider reported nothing,
|
|
19
|
+
* nothing is inserted — there is no byte-derived fallback, because a fabricated
|
|
20
|
+
* count is worse than a missing one.
|
|
21
|
+
*/
|
|
22
|
+
|
|
23
|
+
const { getDb } = require('./db');
|
|
24
|
+
const { normalizeUsage } = require('./token_usage');
|
|
25
|
+
|
|
26
|
+
/**
|
|
27
|
+
* The `source` column enumerates writers, per the approved plan:
|
|
28
|
+
* `gateway` — one bridged provider response. Two writers produce it:
|
|
29
|
+
* `handleAnthropicMessages` (`src/bridge/gateway_adapters.js`) and the
|
|
30
|
+
* per-bridge `/v1/messages` handler (`src/bridge/handlers/anthropic.js`).
|
|
31
|
+
* Both answer the same request shape against the same upstream population, so
|
|
32
|
+
* they share the value; `bridge_name` distinguishes the two.
|
|
33
|
+
* `tool_call` — reserved for mirroring provider usage attached to a
|
|
34
|
+
* `tool_calls` row. Nothing writes it today.
|
|
35
|
+
*/
|
|
36
|
+
const PROVIDER_USAGE_SOURCES = Object.freeze({ GATEWAY: 'gateway', TOOL_CALL: 'tool_call' });
|
|
37
|
+
|
|
38
|
+
function closeQuietly(conn) {
|
|
39
|
+
if (!conn) return;
|
|
40
|
+
try { conn.close(); } catch (_) { /* ignore close error */ }
|
|
41
|
+
}
|
|
42
|
+
|
|
43
|
+
function insertRow(conn, event, measured) {
|
|
44
|
+
conn.prepare(`
|
|
45
|
+
INSERT INTO provider_usage
|
|
46
|
+
(source, provider, model, client, bridge_name, stream,
|
|
47
|
+
input_tokens, output_tokens, cached_tokens, total_tokens)
|
|
48
|
+
VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
|
|
49
|
+
`).run(
|
|
50
|
+
String(event.source || PROVIDER_USAGE_SOURCES.GATEWAY).slice(0, 40),
|
|
51
|
+
event.provider ? String(event.provider).slice(0, 100) : null,
|
|
52
|
+
event.model ? String(event.model).slice(0, 200) : null,
|
|
53
|
+
event.client ? String(event.client).slice(0, 100) : null,
|
|
54
|
+
event.bridgeName ? String(event.bridgeName).slice(0, 100) : null,
|
|
55
|
+
event.stream ? 1 : 0,
|
|
56
|
+
measured.input_tokens,
|
|
57
|
+
measured.output_tokens,
|
|
58
|
+
measured.cache_read_input_tokens,
|
|
59
|
+
measured.total_tokens
|
|
60
|
+
);
|
|
61
|
+
}
|
|
62
|
+
|
|
63
|
+
/**
|
|
64
|
+
* @param {object} event
|
|
65
|
+
* @param {string} event.source - `gateway` | `tool_call`
|
|
66
|
+
* @param {object|null} event.usage - raw provider usage payload
|
|
67
|
+
* @param {string} [event.provider]
|
|
68
|
+
* @param {string} [event.model]
|
|
69
|
+
* @param {string} [event.client]
|
|
70
|
+
* @param {string} [event.bridgeName]
|
|
71
|
+
* @param {boolean} [event.stream]
|
|
72
|
+
* @param {string|null} [dbPath]
|
|
73
|
+
* @returns {{recorded: boolean, reason?: string}}
|
|
74
|
+
*/
|
|
75
|
+
function recordProviderUsage(event = {}, dbPath = null) {
|
|
76
|
+
const measured = normalizeUsage(event.usage);
|
|
77
|
+
if (!measured.available) return { recorded: false, reason: measured.reason };
|
|
78
|
+
|
|
79
|
+
let conn = null;
|
|
80
|
+
try {
|
|
81
|
+
conn = getDb(dbPath || undefined);
|
|
82
|
+
insertRow(conn, event, measured);
|
|
83
|
+
return { recorded: true };
|
|
84
|
+
} catch (_) {
|
|
85
|
+
// Telemetry must never break a response that is already being written.
|
|
86
|
+
return { recorded: false, reason: 'write_failed' };
|
|
87
|
+
} finally {
|
|
88
|
+
closeQuietly(conn);
|
|
89
|
+
}
|
|
90
|
+
}
|
|
91
|
+
|
|
92
|
+
module.exports = { recordProviderUsage, PROVIDER_USAGE_SOURCES };
|
package/src/schema_ddl.js
CHANGED
|
@@ -45,6 +45,21 @@ const CORE_SCHEMA_DDL = `
|
|
|
45
45
|
PRIMARY KEY (client, workspace_root, session_id)
|
|
46
46
|
);
|
|
47
47
|
|
|
48
|
+
CREATE TABLE IF NOT EXISTS provider_usage (
|
|
49
|
+
id INTEGER PRIMARY KEY AUTOINCREMENT,
|
|
50
|
+
timestamp DATETIME DEFAULT CURRENT_TIMESTAMP,
|
|
51
|
+
source TEXT NOT NULL,
|
|
52
|
+
provider TEXT,
|
|
53
|
+
model TEXT,
|
|
54
|
+
client TEXT,
|
|
55
|
+
bridge_name TEXT,
|
|
56
|
+
stream INTEGER NOT NULL DEFAULT 0,
|
|
57
|
+
input_tokens INTEGER,
|
|
58
|
+
output_tokens INTEGER,
|
|
59
|
+
cached_tokens INTEGER,
|
|
60
|
+
total_tokens INTEGER
|
|
61
|
+
);
|
|
62
|
+
|
|
48
63
|
CREATE TABLE IF NOT EXISTS agents (
|
|
49
64
|
name TEXT PRIMARY KEY,
|
|
50
65
|
icon TEXT,
|
|
@@ -69,7 +84,8 @@ const CORE_SCHEMA_DDL = `
|
|
|
69
84
|
provider TEXT NOT NULL,
|
|
70
85
|
enabled INTEGER NOT NULL DEFAULT 1,
|
|
71
86
|
target_url TEXT,
|
|
72
|
-
api_key TEXT
|
|
87
|
+
api_key TEXT,
|
|
88
|
+
supports_stream_usage INTEGER
|
|
73
89
|
);
|
|
74
90
|
|
|
75
91
|
CREATE TABLE IF NOT EXISTS projects (
|
package/src/templates/AGENTS.md
CHANGED
|
@@ -15,7 +15,7 @@
|
|
|
15
15
|
NEVER run destructive commands (`rm -rf`, `DROP`, `git reset`) or view secrets (`.env*`).
|
|
16
16
|
|
|
17
17
|
<!-- KONOHA-CONTRACT-START -->
|
|
18
|
-
## Konoha runtime contract (2.2.
|
|
18
|
+
## Konoha runtime contract (2.2.24)
|
|
19
19
|
main agent on Antigravity IDE/CLI. Mandatory every turn.
|
|
20
20
|
- **Konoha is mandatory**: `find_skill`/`get_skill`; no raw cat/grep.
|
|
21
21
|
- **Code intelligence is mandatory**: `code_search`/`code_symbol`/`code_impact`, absolute root.
|
package/src/templates/GEMINI.md
CHANGED
|
@@ -15,7 +15,7 @@
|
|
|
15
15
|
NEVER run destructive commands (`rm -rf`, `DROP`, `git reset`) or view secrets (`.env*`).
|
|
16
16
|
|
|
17
17
|
<!-- KONOHA-CONTRACT-START -->
|
|
18
|
-
## Konoha runtime contract (2.2.
|
|
18
|
+
## Konoha runtime contract (2.2.24)
|
|
19
19
|
main agent on Antigravity IDE/CLI. Mandatory every turn.
|
|
20
20
|
- **Konoha is mandatory**: `find_skill`/`get_skill`; no raw cat/grep.
|
|
21
21
|
- **Code intelligence is mandatory**: `code_search`/`code_symbol`/`code_impact`, absolute root.
|
|
@@ -36,7 +36,7 @@ In the Konoha workflow, Kage exercises executive technical authority across two
|
|
|
36
36
|
- **Two-Step Zero-AI-Slop Review (TARGET 100/100)**:
|
|
37
37
|
- **Step 1 — `aislop_scan`**: Scans changed files via CLI `rtk aislop scan --changes` (or MCP tool with explicit paths). Zero findings of any severity are permitted.
|
|
38
38
|
- **Step 2 — `anti-slop` Rule Review**: Loads the vendored `antislop` skill via `konoha.get_skill("antislop")` and verifies adherence to craftmanship standards and Delivery Gate rules.
|
|
39
|
-
- **
|
|
39
|
+
- **Supply Chain Dependency Audit Gate**: When maintaining Konoha itself, Kage verifies that `rtk node tests/test_dependency_audit.js` passes: `pnpm audit --json` over the root and `apps/web` workspaces with zero unapproved advisories, no Socket account or token involved. The optional Socket.dev scan (`rtk node scripts/socket_audit_optional.js`, needs `SOCKET_CLI_API_TOKEN`) is advisory only and its `SKIPPED`/`UNVERIFIED` output is never a delivery blocker.
|
|
40
40
|
- **Test Directory Discovery & Single-Directory Invariant**: Explore existing test directories (`tests/`, `test/`, `spec/`) using `find_files_clean`. Always place new tests within the established directory. NEVER create duplicate or conflicting test folders.
|
|
41
41
|
- **Workspace Hygiene & Post-Approval Cleanup**: All temporary debug scripts (`debug_*`, `temp_*`, `test_patch.py`, `scratch/*`) must be removed upon final approval so the repository remains production-clean.
|
|
42
42
|
|