@elizaos/plugin-agent-orchestrator 0.3.2 → 0.3.4
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/dist/index.js
CHANGED
|
@@ -1,12 +1,16 @@
|
|
|
1
|
-
import {createRequire} from "node:module";
|
|
1
|
+
import { createRequire } from "node:module";
|
|
2
2
|
var __defProp = Object.defineProperty;
|
|
3
|
+
var __returnValue = (v) => v;
|
|
4
|
+
function __exportSetter(name, newValue) {
|
|
5
|
+
this[name] = __returnValue.bind(null, newValue);
|
|
6
|
+
}
|
|
3
7
|
var __export = (target, all) => {
|
|
4
8
|
for (var name in all)
|
|
5
9
|
__defProp(target, name, {
|
|
6
10
|
get: all[name],
|
|
7
11
|
enumerable: true,
|
|
8
12
|
configurable: true,
|
|
9
|
-
set: (
|
|
13
|
+
set: __exportSetter.bind(all, name)
|
|
10
14
|
});
|
|
11
15
|
};
|
|
12
16
|
var __esm = (fn, res) => () => (fn && (res = fn(fn = 0)), res);
|
|
@@ -21,7 +25,8 @@ function stripAnsi(raw) {
|
|
|
21
25
|
}
|
|
22
26
|
function cleanForChat(raw) {
|
|
23
27
|
const stripped = applyAnsiStrip(raw);
|
|
24
|
-
return stripped.replace(TUI_DECORATIVE, " ").replace(/\xa0/g, " ").split(
|
|
28
|
+
return stripped.replace(TUI_DECORATIVE, " ").replace(/\xa0/g, " ").split(`
|
|
29
|
+
`).filter((line) => {
|
|
25
30
|
const trimmed = line.trim();
|
|
26
31
|
if (!trimmed)
|
|
27
32
|
return false;
|
|
@@ -32,7 +37,10 @@ function cleanForChat(raw) {
|
|
|
32
37
|
if (!/[a-zA-Z0-9]/.test(trimmed))
|
|
33
38
|
return false;
|
|
34
39
|
return true;
|
|
35
|
-
}).map((line) => line.replace(/ {2,}/g, " ").trim()).filter((line) => line.length > 0).join(
|
|
40
|
+
}).map((line) => line.replace(/ {2,}/g, " ").trim()).filter((line) => line.length > 0).join(`
|
|
41
|
+
`).replace(/\n{3,}/g, `
|
|
42
|
+
|
|
43
|
+
`).trim();
|
|
36
44
|
}
|
|
37
45
|
function extractCompletionSummary(raw) {
|
|
38
46
|
const stripped = applyAnsiStrip(raw);
|
|
@@ -57,7 +65,8 @@ function extractCompletionSummary(raw) {
|
|
|
57
65
|
for (const m of diffStat)
|
|
58
66
|
lines.push(m.trim());
|
|
59
67
|
}
|
|
60
|
-
return lines.join(
|
|
68
|
+
return lines.join(`
|
|
69
|
+
`);
|
|
61
70
|
}
|
|
62
71
|
function extractDevServerUrl(raw) {
|
|
63
72
|
const stripped = applyAnsiStrip(raw);
|
|
@@ -71,7 +80,8 @@ function captureTaskResponse(sessionId, buffers, markers) {
|
|
|
71
80
|
return "";
|
|
72
81
|
const responseLines = buffer.slice(marker);
|
|
73
82
|
markers.delete(sessionId);
|
|
74
|
-
return cleanForChat(responseLines.join(
|
|
83
|
+
return cleanForChat(responseLines.join(`
|
|
84
|
+
`));
|
|
75
85
|
}
|
|
76
86
|
var CURSOR_MOVEMENT, CURSOR_POSITION, ERASE, OSC, ALL_ANSI, CONTROL_CHARS, ORPHAN_SGR, LONG_SPACES, TUI_DECORATIVE, LOADING_LINE, STATUS_LINE;
|
|
77
87
|
var init_ansi_utils = __esm(() => {
|
|
@@ -90,54 +100,208 @@ var init_ansi_utils = __esm(() => {
|
|
|
90
100
|
|
|
91
101
|
// src/services/swarm-coordinator-prompts.ts
|
|
92
102
|
function buildCoordinationPrompt(taskCtx, promptText, recentOutput, decisionHistory) {
|
|
93
|
-
const historySection = decisionHistory.length > 0 ?
|
|
94
|
-
|
|
103
|
+
const historySection = decisionHistory.length > 0 ? `
|
|
104
|
+
Previous decisions for this session:
|
|
105
|
+
${decisionHistory.slice(-5).map((d, i) => ` ${i + 1}. [${d.event}] prompt="${d.promptText}" → ${d.action}${d.response ? ` ("${d.response}")` : ""} — ${d.reasoning}`).join(`
|
|
106
|
+
`)}
|
|
107
|
+
` : "";
|
|
108
|
+
return `You are Milady, an AI orchestrator managing a swarm of coding agents. ` + `A ${taskCtx.agentType} coding agent ("${taskCtx.label}", session: ${taskCtx.sessionId}) ` + `is blocked and waiting for input.
|
|
109
|
+
|
|
110
|
+
` + `Original task: "${taskCtx.originalTask}"
|
|
111
|
+
` + `Working directory: ${taskCtx.workdir}
|
|
112
|
+
` + `Repository: ${taskCtx.repo ?? "none (scratch directory)"}
|
|
113
|
+
` + historySection + `
|
|
114
|
+
Recent terminal output (last 50 lines):
|
|
115
|
+
` + `---
|
|
116
|
+
${recentOutput.slice(-3000)}
|
|
117
|
+
---
|
|
118
|
+
|
|
119
|
+
` + `The agent is showing this blocking prompt:
|
|
120
|
+
` + `"${promptText}"
|
|
121
|
+
|
|
122
|
+
` + `Decide how to respond. Your options:
|
|
123
|
+
|
|
124
|
+
` + `1. "respond" — Send a response to unblock the agent. For text prompts (Y/n, questions), ` + `set "response" to the text to send. For TUI menus or interactive prompts that need ` + `special keys, set "useKeys": true and "keys" to the key sequence ` + `(e.g. ["enter"], ["down","enter"], ["y","enter"]).
|
|
125
|
+
|
|
126
|
+
` + `2. "complete" — The original task has been fulfilled. The agent has finished its work ` + `(e.g. code written, PR created, tests passed) and is back at the idle prompt. ` + `Use this when the terminal output shows the task objectives have been met.
|
|
95
127
|
|
|
96
|
-
` + `
|
|
97
|
-
` + `- When in doubt, escalate \u2014 it's better to ask the human than to make a wrong choice.
|
|
128
|
+
` + `3. "escalate" — The prompt requires human judgment (e.g. design decisions, ` + `ambiguous requirements, security-sensitive actions). Do NOT respond yourself.
|
|
98
129
|
|
|
99
|
-
` + `
|
|
130
|
+
` + `4. "ignore" — The prompt is not actually blocking or is already being handled.
|
|
131
|
+
|
|
132
|
+
` + `Guidelines:
|
|
133
|
+
` + `- IMPORTANT: If the prompt asks to approve access to files or directories OUTSIDE the working ` + `directory (${taskCtx.workdir}), DECLINE the request and REDIRECT the agent. Do NOT approve ` + `access to paths like /etc, ~/.ssh, ~/, /tmp, or any path that doesn't start with the working ` + `directory. Instead, respond with "n" (or the decline option) and tell the agent: ` + `"That path is outside your workspace. Use ${taskCtx.workdir} instead — ` + `create any files or directories you need there." This keeps the agent moving without ` + `granting out-of-scope access. The coordinator will also notify the human in case ` + `broader access was intended.
|
|
134
|
+
` + `- For tool approval prompts (file writes, shell commands, etc.), respond "y" or use keys:["enter"] to approve.
|
|
135
|
+
` + `- For Y/n confirmations that align with the original task, respond "y".
|
|
136
|
+
` + `- For design questions or choices that could go either way, escalate.
|
|
137
|
+
` + `- For error recovery prompts, try to respond if the path forward is clear.
|
|
138
|
+
` + `- If the output shows a PR was just created (e.g. "Created pull request #N"), do NOT use "complete" yet. ` + `Instead respond with "Review your PR, run each test plan item to verify it works, update the PR to check off each item, then confirm all items pass".
|
|
139
|
+
` + `- Only use "complete" if the agent confirmed it verified ALL test plan items after creating the PR.
|
|
140
|
+
` + `- If the agent is asking for information that was NOT provided in the original task ` + `(e.g. which repository to use, project requirements, credentials), ESCALATE. ` + `The coordinator does not have this information — the human must provide it.
|
|
141
|
+
` + `- When in doubt, escalate — it's better to ask the human than to make a wrong choice.
|
|
142
|
+
|
|
143
|
+
` + `Respond with ONLY a JSON object:
|
|
144
|
+
` + `{"action": "respond|complete|escalate|ignore", "response": "...", "useKeys": false, "keys": [], "reasoning": "..."}`;
|
|
100
145
|
}
|
|
101
146
|
function buildIdleCheckPrompt(taskCtx, recentOutput, idleMinutes, idleCheckNumber, maxIdleChecks, decisionHistory) {
|
|
102
|
-
const historySection = decisionHistory.length > 0 ?
|
|
103
|
-
|
|
147
|
+
const historySection = decisionHistory.length > 0 ? `
|
|
148
|
+
Previous decisions for this session:
|
|
149
|
+
${decisionHistory.slice(-5).map((d, i) => ` ${i + 1}. [${d.event}] prompt="${d.promptText}" → ${d.action}${d.response ? ` ("${d.response}")` : ""} — ${d.reasoning}`).join(`
|
|
150
|
+
`)}
|
|
151
|
+
` : "";
|
|
152
|
+
return `You are Milady, an AI orchestrator managing a swarm of coding agents. ` + `A ${taskCtx.agentType} coding agent ("${taskCtx.label}", session: ${taskCtx.sessionId}) ` + `has been idle for ${idleMinutes} minutes with no events or output changes.
|
|
153
|
+
|
|
154
|
+
` + `Original task: "${taskCtx.originalTask}"
|
|
155
|
+
` + `Working directory: ${taskCtx.workdir}
|
|
156
|
+
` + `Repository: ${taskCtx.repo ?? "none (scratch directory)"}
|
|
157
|
+
` + `Idle check: ${idleCheckNumber} of ${maxIdleChecks} (session will be force-escalated after ${maxIdleChecks})
|
|
158
|
+
` + historySection + `
|
|
159
|
+
Recent terminal output (last 50 lines):
|
|
160
|
+
` + `---
|
|
161
|
+
${recentOutput.slice(-3000)}
|
|
162
|
+
---
|
|
163
|
+
|
|
164
|
+
` + `The session has gone silent. Analyze the terminal output and decide:
|
|
165
|
+
|
|
166
|
+
` + `1. "complete" — The task is FULLY done. ALL objectives in the original task were met ` + `AND the final deliverable is visible in the output (e.g. a PR URL was printed, or the ` + `task explicitly did not require a PR). The agent is back at the idle prompt.
|
|
167
|
+
|
|
168
|
+
` + `2. "respond" — The agent appears stuck or waiting for input that wasn't detected ` + `as a blocking prompt. Send a message to nudge it (e.g. "continue", or answer a question ` + `visible in the output). If code was committed but no PR was created yet, respond with ` + `"please create a pull request with your changes" or similar.
|
|
104
169
|
|
|
105
|
-
` + `
|
|
170
|
+
` + `3. "escalate" — Something looks wrong or unclear. The human should review.
|
|
106
171
|
|
|
107
|
-
` + `
|
|
172
|
+
` + `4. "ignore" — The agent is still actively working (e.g. compiling, running tests, ` + `pushing to remote, creating a PR). The idle period is expected and it will produce output soon.
|
|
173
|
+
|
|
174
|
+
` + `Guidelines:
|
|
175
|
+
` + `- IMPORTANT: Do NOT mark "complete" if the original task involves creating a PR and no PR URL ` + `(e.g. github.com/...pull/...) appears in the output. Instead use "respond" to nudge the agent ` + `to create the PR.
|
|
176
|
+
` + `- Do NOT mark "complete" just because code was committed — commits alone don't finish a task ` + `that requires a PR.
|
|
177
|
+
` + `- Network operations (git push, gh pr create, API calls) can cause several minutes of silence — ` + `prefer "ignore" for early idle checks if the agent was mid-workflow.
|
|
178
|
+
` + `- If the output ends with a command prompt ($ or >) and ALL task objectives are confirmed met, use "complete".
|
|
179
|
+
` + `- If the output shows an error or the agent seems stuck in a loop, escalate.
|
|
180
|
+
` + `- If the agent is clearly mid-operation (build output, test runner, git operations), use "ignore".
|
|
181
|
+
` + `- On check ${idleCheckNumber} of ${maxIdleChecks} — if unsure, lean toward "respond" with a nudge rather than "complete".
|
|
182
|
+
|
|
183
|
+
` + `Respond with ONLY a JSON object:
|
|
184
|
+
` + `{"action": "respond|complete|escalate|ignore", "response": "...", "useKeys": false, "keys": [], "reasoning": "..."}`;
|
|
108
185
|
}
|
|
109
186
|
function buildTurnCompletePrompt(taskCtx, turnOutput, decisionHistory) {
|
|
110
|
-
const historySection = decisionHistory.length > 0 ?
|
|
111
|
-
|
|
187
|
+
const historySection = decisionHistory.length > 0 ? `
|
|
188
|
+
Previous decisions for this session:
|
|
189
|
+
${decisionHistory.slice(-5).map((d, i) => ` ${i + 1}. [${d.event}] prompt="${d.promptText}" → ${d.action}${d.response ? ` ("${d.response}")` : ""} — ${d.reasoning}`).join(`
|
|
190
|
+
`)}
|
|
191
|
+
` : "";
|
|
192
|
+
return `You are Milady, an AI orchestrator managing a swarm of coding agents. ` + `A ${taskCtx.agentType} coding agent ("${taskCtx.label}", session: ${taskCtx.sessionId}) ` + `just finished a turn and is back at the idle prompt waiting for input.
|
|
193
|
+
|
|
194
|
+
` + `Original task: "${taskCtx.originalTask}"
|
|
195
|
+
` + `Working directory: ${taskCtx.workdir}
|
|
196
|
+
` + `Repository: ${taskCtx.repo ?? "none (scratch directory)"}
|
|
197
|
+
` + historySection + `
|
|
198
|
+
Output from this turn:
|
|
199
|
+
` + `---
|
|
200
|
+
${turnOutput.slice(-3000)}
|
|
201
|
+
---
|
|
202
|
+
|
|
203
|
+
` + `The agent completed a turn. Decide if the OVERALL task is done or if more work is needed.
|
|
204
|
+
|
|
205
|
+
` + `IMPORTANT: Coding agents work in multiple turns. A single turn completing does NOT mean ` + `the task is done. You must verify that EVERY objective in the original task has been addressed ` + `in the output before declaring "complete".
|
|
206
|
+
|
|
207
|
+
` + `Your options:
|
|
208
|
+
|
|
209
|
+
` + `1. "respond" — The agent finished a step but the overall task is NOT done yet. ` + `Send a follow-up instruction to continue. Set "response" to the next instruction ` + `(e.g. "Now run the tests", "Create a PR with these changes", "Continue with the next part"). ` + `THIS IS THE DEFAULT — most turns are intermediate steps, not the final result.
|
|
112
210
|
|
|
113
|
-
` + `2. "complete"
|
|
211
|
+
` + `2. "complete" — The original task objectives have ALL been fully met. For repo-based tasks, ` + `this means code was written, changes were committed, pushed, AND a pull request was created. ` + `Only use this when you can point to specific evidence in the output for EVERY objective ` + `(e.g. "Created pull request #N" in the output).
|
|
114
212
|
|
|
115
|
-
` + `
|
|
116
|
-
` + `- If the task mentions multiple features/fixes, verify EACH one is addressed, not just the first.\n` + `- If the agent only analyzed code or read files, it hasn't done the actual work yet \u2014 send a follow-up.
|
|
117
|
-
` + `- If the agent wrote code but didn't test it and testing seems appropriate, ask it to run tests.\n` + `- If the output shows errors or failed tests, send a follow-up to fix them.\n` + `- IMPORTANT: If the working directory is a git repository clone (not a scratch dir), the agent ` + `MUST commit its changes, push them, and create a pull request before the task can be "complete". ` + `If the output only shows code edits with no git commit or PR, respond with "Now commit your changes, push, and create a pull request".\n` + `- IMPORTANT: Creating a PR is NOT the final step. If this is the turn where the PR was created ` + `(i.e. "Created pull request" or a PR URL appears for the FIRST time and no previous decision ` + `already sent a review follow-up), respond with "Review your PR, run each test plan item to verify ` + `it works, update the PR to check off each item, then confirm all items pass".\n` + `- If a previous decision ALREADY sent a review/verification follow-up (check the decision history), ` + `and the agent has now responded with its review results, you MAY mark "complete" if the agent ` + `indicates the work is done (e.g. "Done", "verified", "all checks pass", "Here's what I did", ` + `or a clear summary of completed work). Do NOT require exact phrases \u2014 use judgment.
|
|
118
|
-
` + `- Keep follow-up instructions concise and specific.\n` + `- When asking agents to verify work, prefer CLI tools (gh, curl, cat, git diff, etc.) over ` + `browser automation. Browser tools may not be available in headless environments and can cause delays.\n` + `- Default to "respond" \u2014 only use "complete" when you're certain ALL work is done.
|
|
213
|
+
` + `3. "escalate" — Something looks wrong or you're unsure whether the task is complete. ` + `Let the human decide.
|
|
119
214
|
|
|
120
|
-
` + `
|
|
215
|
+
` + `4. "ignore" — Should not normally be used here.
|
|
216
|
+
|
|
217
|
+
` + `Guidelines:
|
|
218
|
+
` + `- BEFORE choosing "complete", enumerate each objective from the original task and verify ` + `evidence in the output. If ANY objective lacks evidence, use "respond" with the missing work.
|
|
219
|
+
` + `- A PR being created does NOT mean the task is done — check that the PR covers ALL requested changes.
|
|
220
|
+
` + `- If the task mentions multiple features/fixes, verify EACH one is addressed, not just the first.
|
|
221
|
+
` + `- If the agent only analyzed code or read files, it hasn't done the actual work yet — send a follow-up.
|
|
222
|
+
` + `- If the agent wrote code but didn't test it and testing seems appropriate, ask it to run tests.
|
|
223
|
+
` + `- If the output shows errors or failed tests, send a follow-up to fix them.
|
|
224
|
+
` + `- IMPORTANT: If the working directory is a git repository clone (not a scratch dir), the agent ` + `MUST commit its changes, push them, and create a pull request before the task can be "complete". ` + `If the output only shows code edits with no git commit or PR, respond with "Now commit your changes, push, and create a pull request".
|
|
225
|
+
` + `- IMPORTANT: Creating a PR is NOT the final step. If this is the turn where the PR was created ` + `(i.e. "Created pull request" or a PR URL appears for the FIRST time and no previous decision ` + `already sent a review follow-up), respond with "Review your PR, run each test plan item to verify ` + `it works, update the PR to check off each item, then confirm all items pass".
|
|
226
|
+
` + `- If a previous decision ALREADY sent a review/verification follow-up (check the decision history), ` + `and the agent has now responded with its review results, you MAY mark "complete" if the agent ` + `indicates the work is done (e.g. "Done", "verified", "all checks pass", "Here's what I did", ` + `or a clear summary of completed work). Do NOT require exact phrases — use judgment.
|
|
227
|
+
` + `- Keep follow-up instructions concise and specific.
|
|
228
|
+
` + `- When asking agents to verify work, prefer CLI tools (gh, curl, cat, git diff, etc.) over ` + `browser automation. Browser tools may not be available in headless environments and can cause delays.
|
|
229
|
+
` + `- Default to "respond" — only use "complete" when you're certain ALL work is done.
|
|
230
|
+
|
|
231
|
+
` + `Respond with ONLY a JSON object:
|
|
232
|
+
` + `{"action": "respond|complete|escalate|ignore", "response": "...", "useKeys": false, "keys": [], "reasoning": "..."}`;
|
|
121
233
|
}
|
|
122
234
|
function buildBlockedEventMessage(taskCtx, promptText, recentOutput, decisionHistory) {
|
|
123
|
-
const historySection = decisionHistory.length > 0 ?
|
|
124
|
-
|
|
125
|
-
` +
|
|
126
|
-
`
|
|
127
|
-
`
|
|
235
|
+
const historySection = decisionHistory.length > 0 ? `
|
|
236
|
+
Previous decisions:
|
|
237
|
+
${decisionHistory.slice(-5).map((d, i) => ` ${i + 1}. [${d.event}] "${d.promptText}" → ${d.action}${d.response ? ` ("${d.response}")` : ""} — ${d.reasoning}`).join(`
|
|
238
|
+
`)}
|
|
239
|
+
` : "";
|
|
240
|
+
return `[Coding Agent Event] A ${taskCtx.agentType} agent ("${taskCtx.label}") is blocked and waiting for input.
|
|
241
|
+
|
|
242
|
+
` + `Task: "${taskCtx.originalTask}"
|
|
243
|
+
` + `Workdir: ${taskCtx.workdir}
|
|
244
|
+
` + `Repo: ${taskCtx.repo ?? "none (scratch directory)"}
|
|
245
|
+
` + historySection + `
|
|
246
|
+
Recent terminal output:
|
|
247
|
+
---
|
|
248
|
+
${recentOutput.slice(-3000)}
|
|
249
|
+
---
|
|
250
|
+
|
|
251
|
+
` + `Blocking prompt: "${promptText}"
|
|
252
|
+
|
|
253
|
+
` + `Decide how to handle this. Options:
|
|
254
|
+
` + `- "respond" — send text or keys to unblock the agent
|
|
255
|
+
` + `- "complete" — the task is fully done
|
|
256
|
+
` + `- "escalate" — you need the user's input
|
|
257
|
+
` + `- "ignore" — not actually blocking
|
|
258
|
+
|
|
259
|
+
` + `Guidelines:
|
|
260
|
+
` + `- For tool approvals / Y/n that align with the task, respond "y" or keys:["enter"].
|
|
261
|
+
` + `- If the prompt asks for info NOT in the original task, escalate.
|
|
262
|
+
` + `- Decline access to paths outside ${taskCtx.workdir}.
|
|
263
|
+
` + `- If a PR was just created, respond to review & verify test plan items before completing.
|
|
264
|
+
` + `- When in doubt, escalate.
|
|
128
265
|
|
|
129
|
-
` + `
|
|
266
|
+
` + `Include a JSON action block at the end of your response:
|
|
267
|
+
` + "```json\n" + `{"action": "respond|complete|escalate|ignore", "response": "...", "useKeys": false, "keys": [], "reasoning": "..."}
|
|
268
|
+
` + "```";
|
|
130
269
|
}
|
|
131
270
|
function buildTurnCompleteEventMessage(taskCtx, turnOutput, decisionHistory) {
|
|
132
|
-
const historySection = decisionHistory.length > 0 ?
|
|
133
|
-
|
|
134
|
-
` +
|
|
135
|
-
`
|
|
136
|
-
`
|
|
271
|
+
const historySection = decisionHistory.length > 0 ? `
|
|
272
|
+
Previous decisions:
|
|
273
|
+
${decisionHistory.slice(-5).map((d, i) => ` ${i + 1}. [${d.event}] "${d.promptText}" → ${d.action}${d.response ? ` ("${d.response}")` : ""} — ${d.reasoning}`).join(`
|
|
274
|
+
`)}
|
|
275
|
+
` : "";
|
|
276
|
+
return `[Coding Agent Event] A ${taskCtx.agentType} agent ("${taskCtx.label}") just finished a turn and is idle.
|
|
137
277
|
|
|
138
|
-
` + `
|
|
278
|
+
` + `Task: "${taskCtx.originalTask}"
|
|
279
|
+
` + `Workdir: ${taskCtx.workdir}
|
|
280
|
+
` + `Repo: ${taskCtx.repo ?? "none (scratch directory)"}
|
|
281
|
+
` + historySection + `
|
|
282
|
+
Turn output:
|
|
283
|
+
---
|
|
284
|
+
${turnOutput.slice(-3000)}
|
|
285
|
+
---
|
|
139
286
|
|
|
140
|
-
` + `
|
|
287
|
+
` + `Decide if the overall task is done or if the agent needs more work.
|
|
288
|
+
|
|
289
|
+
` + `Options:
|
|
290
|
+
` + `- "respond" — send a follow-up instruction (DEFAULT for intermediate steps)
|
|
291
|
+
` + `- "complete" — ALL task objectives met (code written, committed, PR created & verified)
|
|
292
|
+
` + `- "escalate" — something looks wrong, ask the user
|
|
293
|
+
` + `- "ignore" — should not normally be used here
|
|
294
|
+
|
|
295
|
+
` + `Guidelines:
|
|
296
|
+
` + `- Verify evidence for EVERY objective before using "complete".
|
|
297
|
+
` + `- If code was written but not committed/pushed/PR'd, respond with next step.
|
|
298
|
+
` + `- If a PR was just created, respond to review & verify test plan items.
|
|
299
|
+
` + `- When asking agents to verify work, prefer CLI tools (gh, curl, cat, etc.) over browser automation.
|
|
300
|
+
` + `- Default to "respond" — only "complete" when certain ALL work is done.
|
|
301
|
+
|
|
302
|
+
` + `Include a JSON action block at the end of your response:
|
|
303
|
+
` + "```json\n" + `{"action": "respond|complete|escalate|ignore", "response": "...", "useKeys": false, "keys": [], "reasoning": "..."}
|
|
304
|
+
` + "```";
|
|
141
305
|
}
|
|
142
306
|
function parseCoordinationResponse(llmOutput) {
|
|
143
307
|
const jsonMatch = llmOutput.match(/\{[\s\S]*\}/);
|
|
@@ -169,7 +333,7 @@ function parseCoordinationResponse(llmOutput) {
|
|
|
169
333
|
}
|
|
170
334
|
|
|
171
335
|
// src/services/swarm-event-triage.ts
|
|
172
|
-
import {ModelType as ModelType2} from "@elizaos/core";
|
|
336
|
+
import { ModelType as ModelType2 } from "@elizaos/core";
|
|
173
337
|
function classifyByHeuristic(ctx) {
|
|
174
338
|
if (ctx.promptType) {
|
|
175
339
|
if (ROUTINE_PROMPT_TYPES.has(ctx.promptType))
|
|
@@ -196,8 +360,17 @@ function classifyByHeuristic(ctx) {
|
|
|
196
360
|
return null;
|
|
197
361
|
}
|
|
198
362
|
function buildTriagePrompt(ctx) {
|
|
199
|
-
const eventDesc = ctx.eventType === "blocked" ? `BLOCKED prompt: "${ctx.promptText.slice(0, 300)}"` : `TURN COMPLETE. Recent output
|
|
200
|
-
|
|
363
|
+
const eventDesc = ctx.eventType === "blocked" ? `BLOCKED prompt: "${ctx.promptText.slice(0, 300)}"` : `TURN COMPLETE. Recent output:
|
|
364
|
+
${(ctx.recentOutput ?? "").slice(-500)}`;
|
|
365
|
+
return `Classify this coding agent event as "routine" or "creative".
|
|
366
|
+
|
|
367
|
+
` + `Task: ${ctx.originalTask.slice(0, 200)}
|
|
368
|
+
` + `Event: ${eventDesc}
|
|
369
|
+
|
|
370
|
+
` + `"routine" = simple approval, permission, config, yes/no, tool consent, obvious pass/fail.
|
|
371
|
+
` + `"creative" = needs task context, error recovery, design choice, ambiguous situation, approach selection.
|
|
372
|
+
|
|
373
|
+
` + `Respond with ONLY a JSON object: {"tier": "routine"} or {"tier": "creative"}`;
|
|
201
374
|
}
|
|
202
375
|
function parseTriageResponse(llmOutput) {
|
|
203
376
|
const matches = llmOutput.matchAll(/\{[\s\S]*?\}/g);
|
|
@@ -207,15 +380,14 @@ function parseTriageResponse(llmOutput) {
|
|
|
207
380
|
if (parsed.tier === "routine" || parsed.tier === "creative") {
|
|
208
381
|
return parsed.tier;
|
|
209
382
|
}
|
|
210
|
-
} catch {
|
|
211
|
-
}
|
|
383
|
+
} catch {}
|
|
212
384
|
}
|
|
213
385
|
return null;
|
|
214
386
|
}
|
|
215
387
|
async function classifyEventTier(runtime, ctx, log) {
|
|
216
388
|
const heuristicResult = classifyByHeuristic(ctx);
|
|
217
389
|
if (heuristicResult) {
|
|
218
|
-
log(`Triage: heuristic
|
|
390
|
+
log(`Triage: heuristic → ${heuristicResult}`);
|
|
219
391
|
return heuristicResult;
|
|
220
392
|
}
|
|
221
393
|
try {
|
|
@@ -223,12 +395,12 @@ async function classifyEventTier(runtime, ctx, log) {
|
|
|
223
395
|
const result = await runtime.useModel(ModelType2.TEXT_SMALL, { prompt });
|
|
224
396
|
const tier = parseTriageResponse(result);
|
|
225
397
|
if (tier) {
|
|
226
|
-
log(`Triage: LLM
|
|
398
|
+
log(`Triage: LLM → ${tier}`);
|
|
227
399
|
return tier;
|
|
228
400
|
}
|
|
229
|
-
log(`Triage: LLM returned unparseable response
|
|
401
|
+
log(`Triage: LLM returned unparseable response — defaulting to creative`);
|
|
230
402
|
} catch (err) {
|
|
231
|
-
log(`Triage: LLM classifier failed: ${err}
|
|
403
|
+
log(`Triage: LLM classifier failed: ${err} — defaulting to creative`);
|
|
232
404
|
}
|
|
233
405
|
return "creative";
|
|
234
406
|
}
|
|
@@ -298,7 +470,7 @@ __export(exports_swarm_decision_loop, {
|
|
|
298
470
|
checkAllTasksComplete: () => checkAllTasksComplete
|
|
299
471
|
});
|
|
300
472
|
import * as path from "node:path";
|
|
301
|
-
import {ModelType as ModelType3} from "@elizaos/core";
|
|
473
|
+
import { ModelType as ModelType3 } from "@elizaos/core";
|
|
302
474
|
function toContextSummary(taskCtx) {
|
|
303
475
|
return {
|
|
304
476
|
sessionId: taskCtx.sessionId,
|
|
@@ -422,9 +594,10 @@ async function executeDecision(ctx, sessionId, decision) {
|
|
|
422
594
|
try {
|
|
423
595
|
const rawOutput = await ctx.ptyService.getSessionOutput(sessionId, 50);
|
|
424
596
|
summary = extractCompletionSummary(rawOutput);
|
|
425
|
-
} catch {
|
|
426
|
-
}
|
|
427
|
-
|
|
597
|
+
} catch {}
|
|
598
|
+
ctx.sendChatMessage(summary ? `Finished "${taskCtx?.label ?? sessionId}".
|
|
599
|
+
|
|
600
|
+
${summary}` : `Finished "${taskCtx?.label ?? sessionId}".`, "coding-agent");
|
|
428
601
|
ctx.ptyService.stopSession(sessionId, true).catch((err) => {
|
|
429
602
|
ctx.log(`Failed to stop session after LLM-detected completion: ${err}`);
|
|
430
603
|
});
|
|
@@ -541,7 +714,7 @@ async function handleBlocked(ctx, sessionId, taskCtx, data) {
|
|
|
541
714
|
event: "blocked",
|
|
542
715
|
promptText,
|
|
543
716
|
decision: "escalate",
|
|
544
|
-
reasoning: "Supervision level is notify
|
|
717
|
+
reasoning: "Supervision level is notify — broadcasting only"
|
|
545
718
|
});
|
|
546
719
|
break;
|
|
547
720
|
}
|
|
@@ -553,7 +726,7 @@ async function handleTurnComplete(ctx, sessionId, taskCtx, data) {
|
|
|
553
726
|
}
|
|
554
727
|
ctx.inFlightDecisions.add(sessionId);
|
|
555
728
|
try {
|
|
556
|
-
ctx.log(`Turn complete for "${taskCtx.label}"
|
|
729
|
+
ctx.log(`Turn complete for "${taskCtx.label}" — assessing whether task is done`);
|
|
557
730
|
const rawResponse = data.response ?? "";
|
|
558
731
|
let turnOutput = cleanForChat(rawResponse);
|
|
559
732
|
if (!turnOutput) {
|
|
@@ -588,7 +761,7 @@ async function handleTurnComplete(ctx, sessionId, taskCtx, data) {
|
|
|
588
761
|
if (decision)
|
|
589
762
|
decisionFromPipeline = true;
|
|
590
763
|
} catch (err) {
|
|
591
|
-
ctx.log(`Agent decision callback failed for turn-complete: ${err}
|
|
764
|
+
ctx.log(`Agent decision callback failed for turn-complete: ${err} — falling back to small LLM`);
|
|
592
765
|
}
|
|
593
766
|
}
|
|
594
767
|
if (!decision) {
|
|
@@ -604,13 +777,13 @@ async function handleTurnComplete(ctx, sessionId, taskCtx, data) {
|
|
|
604
777
|
}
|
|
605
778
|
}
|
|
606
779
|
if (!decision) {
|
|
607
|
-
ctx.log(`Turn-complete for "${taskCtx.label}": all decision paths failed
|
|
780
|
+
ctx.log(`Turn-complete for "${taskCtx.label}": all decision paths failed — escalating`);
|
|
608
781
|
decision = {
|
|
609
782
|
action: "escalate",
|
|
610
|
-
reasoning: "All decision paths returned invalid response
|
|
783
|
+
reasoning: "All decision paths returned invalid response — escalating for human review"
|
|
611
784
|
};
|
|
612
785
|
}
|
|
613
|
-
ctx.log(`Turn assessment for "${taskCtx.label}": ${decision.action}${decision.action === "respond" ? `
|
|
786
|
+
ctx.log(`Turn assessment for "${taskCtx.label}": ${decision.action}${decision.action === "respond" ? ` → "${(decision.response ?? "").slice(0, 80)}"` : ""} — ${decision.reasoning.slice(0, 120)}`);
|
|
614
787
|
taskCtx.decisions.push({
|
|
615
788
|
timestamp: Date.now(),
|
|
616
789
|
event: "turn_complete",
|
|
@@ -634,7 +807,7 @@ async function handleTurnComplete(ctx, sessionId, taskCtx, data) {
|
|
|
634
807
|
const preview = instruction.length > 120 ? `${instruction.slice(0, 120)}...` : instruction;
|
|
635
808
|
ctx.sendChatMessage(`[${taskCtx.label}] Turn done, continuing: ${preview}`, "coding-agent");
|
|
636
809
|
} else if (decision.action === "escalate") {
|
|
637
|
-
ctx.sendChatMessage(`[${taskCtx.label}] Turn finished
|
|
810
|
+
ctx.sendChatMessage(`[${taskCtx.label}] Turn finished — needs your attention: ${decision.reasoning}`, "coding-agent");
|
|
638
811
|
}
|
|
639
812
|
}
|
|
640
813
|
await executeDecision(ctx, sessionId, decision);
|
|
@@ -674,7 +847,7 @@ async function handleAutonomousDecision(ctx, sessionId, taskCtx, promptText, rec
|
|
|
674
847
|
if (decision)
|
|
675
848
|
decisionFromPipeline = true;
|
|
676
849
|
} catch (err) {
|
|
677
|
-
ctx.log(`Agent decision callback failed: ${err}
|
|
850
|
+
ctx.log(`Agent decision callback failed: ${err} — falling back to small LLM`);
|
|
678
851
|
}
|
|
679
852
|
}
|
|
680
853
|
if (!decision) {
|
|
@@ -703,7 +876,7 @@ async function handleAutonomousDecision(ctx, sessionId, taskCtx, promptText, rec
|
|
|
703
876
|
if (decision.action === "respond" && isOutOfScopeAccess(promptText, taskCtx.workdir)) {
|
|
704
877
|
decision = {
|
|
705
878
|
action: "respond",
|
|
706
|
-
response: `No
|
|
879
|
+
response: `No — that path is outside your workspace. Use ${taskCtx.workdir} instead. Create any files or directories you need there.`,
|
|
707
880
|
reasoning: `Declined out-of-scope access (outside ${taskCtx.workdir}) and redirected agent to workspace.`
|
|
708
881
|
};
|
|
709
882
|
ctx.sendChatMessage(`[${taskCtx.label}] Declined out-of-scope access and redirected to workspace (${taskCtx.workdir}). If you intended broader access, send the agent an override.`, "coding-agent");
|
|
@@ -733,7 +906,7 @@ async function handleAutonomousDecision(ctx, sessionId, taskCtx, promptText, rec
|
|
|
733
906
|
if (decision.action === "respond") {
|
|
734
907
|
const actionDesc = decision.useKeys ? `Sent keys: ${decision.keys?.join(", ")}` : decision.response ? `Responded: ${decision.response.length > 100 ? `${decision.response.slice(0, 100)}...` : decision.response}` : "Responded";
|
|
735
908
|
const reasonExcerpt = decision.reasoning.length > 150 ? `${decision.reasoning.slice(0, 150)}...` : decision.reasoning;
|
|
736
|
-
ctx.sendChatMessage(`[${taskCtx.label}] ${actionDesc}
|
|
909
|
+
ctx.sendChatMessage(`[${taskCtx.label}] ${actionDesc} — ${reasonExcerpt}`, "coding-agent");
|
|
737
910
|
} else if (decision.action === "escalate") {
|
|
738
911
|
ctx.sendChatMessage(`[${taskCtx.label}] Needs your attention: ${decision.reasoning}`, "coding-agent");
|
|
739
912
|
}
|
|
@@ -773,7 +946,7 @@ async function handleConfirmDecision(ctx, sessionId, taskCtx, promptText, recent
|
|
|
773
946
|
if (decision)
|
|
774
947
|
decisionFromPipeline = true;
|
|
775
948
|
} catch (err) {
|
|
776
|
-
ctx.log(`Agent decision callback failed (confirm): ${err}
|
|
949
|
+
ctx.log(`Agent decision callback failed (confirm): ${err} — falling back to small LLM`);
|
|
777
950
|
}
|
|
778
951
|
}
|
|
779
952
|
if (!decision) {
|
|
@@ -787,7 +960,7 @@ async function handleConfirmDecision(ctx, sessionId, taskCtx, promptText, recent
|
|
|
787
960
|
recentOutput: output,
|
|
788
961
|
llmDecision: {
|
|
789
962
|
action: "escalate",
|
|
790
|
-
reasoning: "All decision paths returned invalid response
|
|
963
|
+
reasoning: "All decision paths returned invalid response — needs human review"
|
|
791
964
|
},
|
|
792
965
|
taskContext: taskCtx,
|
|
793
966
|
createdAt: Date.now()
|
|
@@ -911,7 +1084,9 @@ var finalizeWorkspaceAction = {
|
|
|
911
1084
|
data: { workspaceId, status }
|
|
912
1085
|
};
|
|
913
1086
|
}
|
|
914
|
-
const commitMessage = content.commitMessage ?? `feat: automated changes from coding agent
|
|
1087
|
+
const commitMessage = content.commitMessage ?? `feat: automated changes from coding agent
|
|
1088
|
+
|
|
1089
|
+
Generated by Milady coding agent plugin.`;
|
|
915
1090
|
const commitHash = await workspaceService.commit(workspaceId, {
|
|
916
1091
|
message: commitMessage,
|
|
917
1092
|
all: true
|
|
@@ -920,7 +1095,15 @@ var finalizeWorkspaceAction = {
|
|
|
920
1095
|
let prInfo = null;
|
|
921
1096
|
if (!content.skipPR) {
|
|
922
1097
|
const prTitle = content.prTitle ?? `[Milady] ${workspace.branch}`;
|
|
923
|
-
const prBody = content.prBody ?? `## Summary
|
|
1098
|
+
const prBody = content.prBody ?? `## Summary
|
|
1099
|
+
|
|
1100
|
+
Automated changes generated by Milady coding agent.
|
|
1101
|
+
|
|
1102
|
+
` + `**Branch:** ${workspace.branch}
|
|
1103
|
+
` + `**Commit:** ${commitHash}
|
|
1104
|
+
|
|
1105
|
+
` + `---
|
|
1106
|
+
*Generated by @elizaos/plugin-agent-orchestrator*`;
|
|
924
1107
|
prInfo = await workspaceService.createPR(workspaceId, {
|
|
925
1108
|
title: prTitle,
|
|
926
1109
|
body: prBody,
|
|
@@ -931,11 +1114,14 @@ var finalizeWorkspaceAction = {
|
|
|
931
1114
|
if (callback) {
|
|
932
1115
|
if (prInfo) {
|
|
933
1116
|
await callback({
|
|
934
|
-
text: `Workspace finalized
|
|
1117
|
+
text: `Workspace finalized!
|
|
1118
|
+
` + `Commit: ${commitHash.slice(0, 8)}
|
|
1119
|
+
` + `PR #${prInfo.number}: ${prInfo.url}`
|
|
935
1120
|
});
|
|
936
1121
|
} else {
|
|
937
1122
|
await callback({
|
|
938
|
-
text: `Workspace changes committed and pushed
|
|
1123
|
+
text: `Workspace changes committed and pushed.
|
|
1124
|
+
` + `Commit: ${commitHash.slice(0, 8)}`
|
|
939
1125
|
});
|
|
940
1126
|
}
|
|
941
1127
|
}
|
|
@@ -1079,18 +1265,23 @@ var listAgentsAction = {
|
|
|
1079
1265
|
}));
|
|
1080
1266
|
const lines = sessions.map((session, index) => {
|
|
1081
1267
|
const statusEmoji = {
|
|
1082
|
-
running: "
|
|
1083
|
-
idle: "
|
|
1084
|
-
blocked: "
|
|
1085
|
-
completed: "
|
|
1086
|
-
error: "
|
|
1087
|
-
}[session.status] ?? "
|
|
1268
|
+
running: "▶️",
|
|
1269
|
+
idle: "⏸️",
|
|
1270
|
+
blocked: "⚠️",
|
|
1271
|
+
completed: "✅",
|
|
1272
|
+
error: "❌"
|
|
1273
|
+
}[session.status] ?? "❓";
|
|
1088
1274
|
return `${index + 1}. ${statusEmoji} ${session.agentType} (${session.id.slice(0, 8)}...)
|
|
1089
|
-
\uD83D\uDCC1 ${session.workdir}
|
|
1275
|
+
\uD83D\uDCC1 ${session.workdir}
|
|
1276
|
+
Status: ${session.status}`;
|
|
1090
1277
|
});
|
|
1091
1278
|
if (callback) {
|
|
1092
1279
|
await callback({
|
|
1093
|
-
text: `Active coding agents
|
|
1280
|
+
text: `Active coding agents:
|
|
1281
|
+
|
|
1282
|
+
${lines.join(`
|
|
1283
|
+
|
|
1284
|
+
`)}`
|
|
1094
1285
|
});
|
|
1095
1286
|
}
|
|
1096
1287
|
return {
|
|
@@ -1103,6 +1294,150 @@ var listAgentsAction = {
|
|
|
1103
1294
|
};
|
|
1104
1295
|
|
|
1105
1296
|
// src/actions/manage-issues.ts
|
|
1297
|
+
var manageIssuesAction = {
|
|
1298
|
+
name: "MANAGE_ISSUES",
|
|
1299
|
+
similes: [
|
|
1300
|
+
"CREATE_ISSUE",
|
|
1301
|
+
"LIST_ISSUES",
|
|
1302
|
+
"CLOSE_ISSUE",
|
|
1303
|
+
"COMMENT_ISSUE",
|
|
1304
|
+
"UPDATE_ISSUE",
|
|
1305
|
+
"GET_ISSUE"
|
|
1306
|
+
],
|
|
1307
|
+
description: "Manage GitHub issues for a repository. " + "Supports creating issues, listing issues, getting issue details, " + "adding comments, updating, closing, and reopening issues.",
|
|
1308
|
+
examples: [
|
|
1309
|
+
[
|
|
1310
|
+
{
|
|
1311
|
+
name: "{{user1}}",
|
|
1312
|
+
content: {
|
|
1313
|
+
text: "Create an issue on the testbed repo to add a login page"
|
|
1314
|
+
}
|
|
1315
|
+
},
|
|
1316
|
+
{
|
|
1317
|
+
name: "{{agentName}}",
|
|
1318
|
+
content: {
|
|
1319
|
+
text: "I'll create that issue for you.",
|
|
1320
|
+
action: "MANAGE_ISSUES"
|
|
1321
|
+
}
|
|
1322
|
+
}
|
|
1323
|
+
],
|
|
1324
|
+
[
|
|
1325
|
+
{
|
|
1326
|
+
name: "{{user1}}",
|
|
1327
|
+
content: {
|
|
1328
|
+
text: "List the open issues on HaruHunab1320/git-workspace-service-testbed"
|
|
1329
|
+
}
|
|
1330
|
+
},
|
|
1331
|
+
{
|
|
1332
|
+
name: "{{agentName}}",
|
|
1333
|
+
content: {
|
|
1334
|
+
text: "Let me check the open issues for that repo.",
|
|
1335
|
+
action: "MANAGE_ISSUES"
|
|
1336
|
+
}
|
|
1337
|
+
}
|
|
1338
|
+
],
|
|
1339
|
+
[
|
|
1340
|
+
{
|
|
1341
|
+
name: "{{user1}}",
|
|
1342
|
+
content: { text: "Close issue #3 on the testbed repo" }
|
|
1343
|
+
},
|
|
1344
|
+
{
|
|
1345
|
+
name: "{{agentName}}",
|
|
1346
|
+
content: {
|
|
1347
|
+
text: "I'll close that issue.",
|
|
1348
|
+
action: "MANAGE_ISSUES"
|
|
1349
|
+
}
|
|
1350
|
+
}
|
|
1351
|
+
]
|
|
1352
|
+
],
|
|
1353
|
+
validate: async (runtime, _message) => {
|
|
1354
|
+
const workspaceService = runtime.getService("CODING_WORKSPACE_SERVICE");
|
|
1355
|
+
return workspaceService != null;
|
|
1356
|
+
},
|
|
1357
|
+
handler: async (runtime, message, _state, options, callback) => {
|
|
1358
|
+
const workspaceService = runtime.getService("CODING_WORKSPACE_SERVICE");
|
|
1359
|
+
if (!workspaceService) {
|
|
1360
|
+
if (callback) {
|
|
1361
|
+
await callback({ text: "Workspace Service is not available." });
|
|
1362
|
+
}
|
|
1363
|
+
return { success: false, error: "SERVICE_UNAVAILABLE" };
|
|
1364
|
+
}
|
|
1365
|
+
workspaceService.setAuthPromptCallback((prompt) => {
|
|
1366
|
+
if (callback) {
|
|
1367
|
+
callback({
|
|
1368
|
+
text: `I need GitHub access to manage issues. Please authorize me:
|
|
1369
|
+
|
|
1370
|
+
` + `Go to: ${prompt.verificationUri}
|
|
1371
|
+
` + `Enter code: **${prompt.userCode}**
|
|
1372
|
+
|
|
1373
|
+
` + `This code expires in ${Math.floor(prompt.expiresIn / 60)} minutes. ` + `I'll wait for you to complete authorization...`
|
|
1374
|
+
});
|
|
1375
|
+
}
|
|
1376
|
+
});
|
|
1377
|
+
const params = options?.parameters;
|
|
1378
|
+
const content = message.content;
|
|
1379
|
+
const text = content.text ?? "";
|
|
1380
|
+
const operation = params?.operation ?? content.operation ?? inferOperation(text);
|
|
1381
|
+
const repo = params?.repo ?? content.repo;
|
|
1382
|
+
if (!repo) {
|
|
1383
|
+
const urlMatch = text?.match(/(?:https?:\/\/github\.com\/)?([a-zA-Z0-9_.-]+\/[a-zA-Z0-9_.-]+)/);
|
|
1384
|
+
if (!urlMatch) {
|
|
1385
|
+
if (callback) {
|
|
1386
|
+
await callback({
|
|
1387
|
+
text: "Please specify a repository (e.g., owner/repo or a GitHub URL)."
|
|
1388
|
+
});
|
|
1389
|
+
}
|
|
1390
|
+
return { success: false, error: "MISSING_REPO" };
|
|
1391
|
+
}
|
|
1392
|
+
return handleOperation(workspaceService, urlMatch[1], operation, params ?? content, text, callback);
|
|
1393
|
+
}
|
|
1394
|
+
return handleOperation(workspaceService, repo, operation, params ?? content, text, callback);
|
|
1395
|
+
},
|
|
1396
|
+
parameters: [
|
|
1397
|
+
{
|
|
1398
|
+
name: "operation",
|
|
1399
|
+
description: "The operation to perform: create, list, get, update, comment, close, reopen, add_labels",
|
|
1400
|
+
required: true,
|
|
1401
|
+
schema: { type: "string" }
|
|
1402
|
+
},
|
|
1403
|
+
{
|
|
1404
|
+
name: "repo",
|
|
1405
|
+
description: "Repository in owner/repo format or full GitHub URL.",
|
|
1406
|
+
required: true,
|
|
1407
|
+
schema: { type: "string" }
|
|
1408
|
+
},
|
|
1409
|
+
{
|
|
1410
|
+
name: "title",
|
|
1411
|
+
description: "Issue title (for create operation).",
|
|
1412
|
+
required: false,
|
|
1413
|
+
schema: { type: "string" }
|
|
1414
|
+
},
|
|
1415
|
+
{
|
|
1416
|
+
name: "body",
|
|
1417
|
+
description: "Issue body/description (for create or comment operations).",
|
|
1418
|
+
required: false,
|
|
1419
|
+
schema: { type: "string" }
|
|
1420
|
+
},
|
|
1421
|
+
{
|
|
1422
|
+
name: "issueNumber",
|
|
1423
|
+
description: "Issue number (for get, update, comment, close, reopen operations).",
|
|
1424
|
+
required: false,
|
|
1425
|
+
schema: { type: "number" }
|
|
1426
|
+
},
|
|
1427
|
+
{
|
|
1428
|
+
name: "labels",
|
|
1429
|
+
description: "Labels to add (comma-separated string or array).",
|
|
1430
|
+
required: false,
|
|
1431
|
+
schema: { type: "string" }
|
|
1432
|
+
},
|
|
1433
|
+
{
|
|
1434
|
+
name: "state",
|
|
1435
|
+
description: "Filter by state: open, closed, or all (for list operation).",
|
|
1436
|
+
required: false,
|
|
1437
|
+
schema: { type: "string" }
|
|
1438
|
+
}
|
|
1439
|
+
]
|
|
1440
|
+
};
|
|
1106
1441
|
async function handleOperation(service, repo, operation, params, originalText, callback) {
|
|
1107
1442
|
try {
|
|
1108
1443
|
switch (operation.toLowerCase()) {
|
|
@@ -1123,9 +1458,12 @@ async function handleOperation(service, repo, operation, params, originalText, c
|
|
|
1123
1458
|
created.push(issue2);
|
|
1124
1459
|
}
|
|
1125
1460
|
if (callback) {
|
|
1126
|
-
const summary = created.map((i) => `#${i.number}: ${i.title}
|
|
1461
|
+
const summary = created.map((i) => `#${i.number}: ${i.title}
|
|
1462
|
+
${i.url}`).join(`
|
|
1463
|
+
`);
|
|
1127
1464
|
await callback({
|
|
1128
|
-
text: `Created ${created.length} issues
|
|
1465
|
+
text: `Created ${created.length} issues:
|
|
1466
|
+
${summary}`
|
|
1129
1467
|
});
|
|
1130
1468
|
}
|
|
1131
1469
|
return { success: true, data: { issues: created } };
|
|
@@ -1142,7 +1480,8 @@ async function handleOperation(service, repo, operation, params, originalText, c
|
|
|
1142
1480
|
});
|
|
1143
1481
|
if (callback) {
|
|
1144
1482
|
await callback({
|
|
1145
|
-
text: `Created issue #${issue.number}: ${issue.title}
|
|
1483
|
+
text: `Created issue #${issue.number}: ${issue.title}
|
|
1484
|
+
${issue.url}`
|
|
1146
1485
|
});
|
|
1147
1486
|
}
|
|
1148
1487
|
return { success: true, data: { issue } };
|
|
@@ -1160,8 +1499,10 @@ async function handleOperation(service, repo, operation, params, originalText, c
|
|
|
1160
1499
|
text: `No ${stateFilter} issues found in ${repo}.`
|
|
1161
1500
|
});
|
|
1162
1501
|
} else {
|
|
1163
|
-
const summary = issues.map((i) => `#${i.number} [${i.state}] ${i.title}${i.labels.length > 0 ? ` (${i.labels.join(", ")})` : ""}`).join(
|
|
1164
|
-
|
|
1502
|
+
const summary = issues.map((i) => `#${i.number} [${i.state}] ${i.title}${i.labels.length > 0 ? ` (${i.labels.join(", ")})` : ""}`).join(`
|
|
1503
|
+
`);
|
|
1504
|
+
await callback({ text: `Issues in ${repo}:
|
|
1505
|
+
${summary}` });
|
|
1165
1506
|
}
|
|
1166
1507
|
}
|
|
1167
1508
|
return { success: true, data: { issues } };
|
|
@@ -1176,7 +1517,12 @@ async function handleOperation(service, repo, operation, params, originalText, c
|
|
|
1176
1517
|
const issue = await service.getIssue(repo, issueNumber);
|
|
1177
1518
|
if (callback) {
|
|
1178
1519
|
await callback({
|
|
1179
|
-
text: `Issue #${issue.number}: ${issue.title} [${issue.state}]
|
|
1520
|
+
text: `Issue #${issue.number}: ${issue.title} [${issue.state}]
|
|
1521
|
+
|
|
1522
|
+
${issue.body}
|
|
1523
|
+
|
|
1524
|
+
Labels: ${issue.labels.join(", ") || "none"}
|
|
1525
|
+
${issue.url}`
|
|
1180
1526
|
});
|
|
1181
1527
|
}
|
|
1182
1528
|
return { success: true, data: { issue } };
|
|
@@ -1345,145 +1691,6 @@ function parseLabels(input) {
|
|
|
1345
1691
|
return input.split(",").map((s) => s.trim()).filter(Boolean);
|
|
1346
1692
|
return [];
|
|
1347
1693
|
}
|
|
1348
|
-
var manageIssuesAction = {
|
|
1349
|
-
name: "MANAGE_ISSUES",
|
|
1350
|
-
similes: [
|
|
1351
|
-
"CREATE_ISSUE",
|
|
1352
|
-
"LIST_ISSUES",
|
|
1353
|
-
"CLOSE_ISSUE",
|
|
1354
|
-
"COMMENT_ISSUE",
|
|
1355
|
-
"UPDATE_ISSUE",
|
|
1356
|
-
"GET_ISSUE"
|
|
1357
|
-
],
|
|
1358
|
-
description: "Manage GitHub issues for a repository. " + "Supports creating issues, listing issues, getting issue details, " + "adding comments, updating, closing, and reopening issues.",
|
|
1359
|
-
examples: [
|
|
1360
|
-
[
|
|
1361
|
-
{
|
|
1362
|
-
name: "{{user1}}",
|
|
1363
|
-
content: {
|
|
1364
|
-
text: "Create an issue on the testbed repo to add a login page"
|
|
1365
|
-
}
|
|
1366
|
-
},
|
|
1367
|
-
{
|
|
1368
|
-
name: "{{agentName}}",
|
|
1369
|
-
content: {
|
|
1370
|
-
text: "I'll create that issue for you.",
|
|
1371
|
-
action: "MANAGE_ISSUES"
|
|
1372
|
-
}
|
|
1373
|
-
}
|
|
1374
|
-
],
|
|
1375
|
-
[
|
|
1376
|
-
{
|
|
1377
|
-
name: "{{user1}}",
|
|
1378
|
-
content: {
|
|
1379
|
-
text: "List the open issues on HaruHunab1320/git-workspace-service-testbed"
|
|
1380
|
-
}
|
|
1381
|
-
},
|
|
1382
|
-
{
|
|
1383
|
-
name: "{{agentName}}",
|
|
1384
|
-
content: {
|
|
1385
|
-
text: "Let me check the open issues for that repo.",
|
|
1386
|
-
action: "MANAGE_ISSUES"
|
|
1387
|
-
}
|
|
1388
|
-
}
|
|
1389
|
-
],
|
|
1390
|
-
[
|
|
1391
|
-
{
|
|
1392
|
-
name: "{{user1}}",
|
|
1393
|
-
content: { text: "Close issue #3 on the testbed repo" }
|
|
1394
|
-
},
|
|
1395
|
-
{
|
|
1396
|
-
name: "{{agentName}}",
|
|
1397
|
-
content: {
|
|
1398
|
-
text: "I'll close that issue.",
|
|
1399
|
-
action: "MANAGE_ISSUES"
|
|
1400
|
-
}
|
|
1401
|
-
}
|
|
1402
|
-
]
|
|
1403
|
-
],
|
|
1404
|
-
validate: async (runtime, _message) => {
|
|
1405
|
-
const workspaceService = runtime.getService("CODING_WORKSPACE_SERVICE");
|
|
1406
|
-
return workspaceService != null;
|
|
1407
|
-
},
|
|
1408
|
-
handler: async (runtime, message, _state, options, callback) => {
|
|
1409
|
-
const workspaceService = runtime.getService("CODING_WORKSPACE_SERVICE");
|
|
1410
|
-
if (!workspaceService) {
|
|
1411
|
-
if (callback) {
|
|
1412
|
-
await callback({ text: "Workspace Service is not available." });
|
|
1413
|
-
}
|
|
1414
|
-
return { success: false, error: "SERVICE_UNAVAILABLE" };
|
|
1415
|
-
}
|
|
1416
|
-
workspaceService.setAuthPromptCallback((prompt) => {
|
|
1417
|
-
if (callback) {
|
|
1418
|
-
callback({
|
|
1419
|
-
text: `I need GitHub access to manage issues. Please authorize me:\n\n` + `Go to: ${prompt.verificationUri}\n` + `Enter code: **${prompt.userCode}**\n\n` + `This code expires in ${Math.floor(prompt.expiresIn / 60)} minutes. ` + `I'll wait for you to complete authorization...`
|
|
1420
|
-
});
|
|
1421
|
-
}
|
|
1422
|
-
});
|
|
1423
|
-
const params = options?.parameters;
|
|
1424
|
-
const content = message.content;
|
|
1425
|
-
const text = content.text ?? "";
|
|
1426
|
-
const operation = params?.operation ?? content.operation ?? inferOperation(text);
|
|
1427
|
-
const repo = params?.repo ?? content.repo;
|
|
1428
|
-
if (!repo) {
|
|
1429
|
-
const urlMatch = text?.match(/(?:https?:\/\/github\.com\/)?([a-zA-Z0-9_.-]+\/[a-zA-Z0-9_.-]+)/);
|
|
1430
|
-
if (!urlMatch) {
|
|
1431
|
-
if (callback) {
|
|
1432
|
-
await callback({
|
|
1433
|
-
text: "Please specify a repository (e.g., owner/repo or a GitHub URL)."
|
|
1434
|
-
});
|
|
1435
|
-
}
|
|
1436
|
-
return { success: false, error: "MISSING_REPO" };
|
|
1437
|
-
}
|
|
1438
|
-
return handleOperation(workspaceService, urlMatch[1], operation, params ?? content, text, callback);
|
|
1439
|
-
}
|
|
1440
|
-
return handleOperation(workspaceService, repo, operation, params ?? content, text, callback);
|
|
1441
|
-
},
|
|
1442
|
-
parameters: [
|
|
1443
|
-
{
|
|
1444
|
-
name: "operation",
|
|
1445
|
-
description: "The operation to perform: create, list, get, update, comment, close, reopen, add_labels",
|
|
1446
|
-
required: true,
|
|
1447
|
-
schema: { type: "string" }
|
|
1448
|
-
},
|
|
1449
|
-
{
|
|
1450
|
-
name: "repo",
|
|
1451
|
-
description: "Repository in owner/repo format or full GitHub URL.",
|
|
1452
|
-
required: true,
|
|
1453
|
-
schema: { type: "string" }
|
|
1454
|
-
},
|
|
1455
|
-
{
|
|
1456
|
-
name: "title",
|
|
1457
|
-
description: "Issue title (for create operation).",
|
|
1458
|
-
required: false,
|
|
1459
|
-
schema: { type: "string" }
|
|
1460
|
-
},
|
|
1461
|
-
{
|
|
1462
|
-
name: "body",
|
|
1463
|
-
description: "Issue body/description (for create or comment operations).",
|
|
1464
|
-
required: false,
|
|
1465
|
-
schema: { type: "string" }
|
|
1466
|
-
},
|
|
1467
|
-
{
|
|
1468
|
-
name: "issueNumber",
|
|
1469
|
-
description: "Issue number (for get, update, comment, close, reopen operations).",
|
|
1470
|
-
required: false,
|
|
1471
|
-
schema: { type: "number" }
|
|
1472
|
-
},
|
|
1473
|
-
{
|
|
1474
|
-
name: "labels",
|
|
1475
|
-
description: "Labels to add (comma-separated string or array).",
|
|
1476
|
-
required: false,
|
|
1477
|
-
schema: { type: "string" }
|
|
1478
|
-
},
|
|
1479
|
-
{
|
|
1480
|
-
name: "state",
|
|
1481
|
-
description: "Filter by state: open, closed, or all (for list operation).",
|
|
1482
|
-
required: false,
|
|
1483
|
-
schema: { type: "string" }
|
|
1484
|
-
}
|
|
1485
|
-
]
|
|
1486
|
-
};
|
|
1487
1694
|
|
|
1488
1695
|
// src/actions/provision-workspace.ts
|
|
1489
1696
|
var provisionWorkspaceAction = {
|
|
@@ -1596,7 +1803,9 @@ var provisionWorkspaceAction = {
|
|
|
1596
1803
|
}
|
|
1597
1804
|
if (callback) {
|
|
1598
1805
|
await callback({
|
|
1599
|
-
text: `Created workspace at ${workspace.path}
|
|
1806
|
+
text: `Created workspace at ${workspace.path}
|
|
1807
|
+
` + `Branch: ${workspace.branch}
|
|
1808
|
+
` + `Type: ${workspace.isWorktree ? "worktree" : "clone"}`
|
|
1600
1809
|
});
|
|
1601
1810
|
}
|
|
1602
1811
|
return {
|
|
@@ -1805,19 +2014,19 @@ var sendToAgentAction = {
|
|
|
1805
2014
|
import * as os from "node:os";
|
|
1806
2015
|
import * as path2 from "node:path";
|
|
1807
2016
|
import {
|
|
1808
|
-
logger as logger3
|
|
2017
|
+
logger as logger3
|
|
1809
2018
|
} from "@elizaos/core";
|
|
1810
2019
|
|
|
1811
2020
|
// src/services/pty-service.ts
|
|
1812
|
-
import {mkdir, readFile, writeFile} from "node:fs/promises";
|
|
1813
|
-
import {dirname, join} from "node:path";
|
|
1814
|
-
import {logger as logger2} from "@elizaos/core";
|
|
2021
|
+
import { mkdir, readFile, writeFile } from "node:fs/promises";
|
|
2022
|
+
import { dirname, join } from "node:path";
|
|
2023
|
+
import { logger as logger2 } from "@elizaos/core";
|
|
1815
2024
|
import {
|
|
1816
|
-
checkAdapters,
|
|
1817
|
-
createAdapter,
|
|
1818
|
-
generateApprovalConfig
|
|
2025
|
+
checkAdapters,
|
|
2026
|
+
createAdapter,
|
|
2027
|
+
generateApprovalConfig
|
|
1819
2028
|
} from "coding-agent-adapters";
|
|
1820
|
-
import {PTYConsoleBridge} from "pty-console";
|
|
2029
|
+
import { PTYConsoleBridge } from "pty-console";
|
|
1821
2030
|
|
|
1822
2031
|
// src/services/agent-metrics.ts
|
|
1823
2032
|
class AgentMetricsTracker {
|
|
@@ -1873,6 +2082,7 @@ function computeAgentScore(metrics) {
|
|
|
1873
2082
|
const speedPenalty = Math.min(avgCompletionMs / 300000, 1) * 0.1;
|
|
1874
2083
|
return Math.max(0, successRate - stallPenalty - speedPenalty);
|
|
1875
2084
|
}
|
|
2085
|
+
var DEFAULT_ORDER = ["claude", "gemini", "codex", "aider"];
|
|
1876
2086
|
function selectAgentType(ctx) {
|
|
1877
2087
|
if (ctx.config.strategy === "fixed") {
|
|
1878
2088
|
return ctx.config.fixedAgentType;
|
|
@@ -1894,7 +2104,6 @@ function selectAgentType(ctx) {
|
|
|
1894
2104
|
}
|
|
1895
2105
|
return bestAgent;
|
|
1896
2106
|
}
|
|
1897
|
-
var DEFAULT_ORDER = ["claude", "gemini", "codex", "aider"];
|
|
1898
2107
|
|
|
1899
2108
|
// src/services/pty-auto-response.ts
|
|
1900
2109
|
async function pushDefaultRules(ctx, sessionId, agentType) {
|
|
@@ -1972,20 +2181,25 @@ async function handleGeminiAuth(ctx, sessionId, sendKeysToSession) {
|
|
|
1972
2181
|
|
|
1973
2182
|
// src/services/pty-init.ts
|
|
1974
2183
|
init_ansi_utils();
|
|
1975
|
-
import {createRequire as createRequire2} from "node:module";
|
|
1976
|
-
import {createAllAdapters} from "coding-agent-adapters";
|
|
2184
|
+
import { createRequire as createRequire2 } from "node:module";
|
|
2185
|
+
import { createAllAdapters } from "coding-agent-adapters";
|
|
1977
2186
|
import {
|
|
1978
|
-
BunCompatiblePTYManager,
|
|
1979
|
-
isBun,
|
|
1980
|
-
PTYManager,
|
|
1981
|
-
ShellAdapter
|
|
2187
|
+
BunCompatiblePTYManager,
|
|
2188
|
+
isBun,
|
|
2189
|
+
PTYManager,
|
|
2190
|
+
ShellAdapter
|
|
1982
2191
|
} from "pty-manager";
|
|
2192
|
+
var _require = createRequire2(import.meta.url);
|
|
2193
|
+
var resolvedAdapterModule = "coding-agent-adapters";
|
|
2194
|
+
try {
|
|
2195
|
+
resolvedAdapterModule = _require.resolve("coding-agent-adapters");
|
|
2196
|
+
} catch {}
|
|
1983
2197
|
function forwardReadyAsTaskComplete(ctx, session) {
|
|
1984
2198
|
if (!ctx.hasActiveTask?.(session.id) || !ctx.hasTaskActivity?.(session.id)) {
|
|
1985
2199
|
return;
|
|
1986
2200
|
}
|
|
1987
2201
|
const response = ctx.taskResponseMarkers.has(session.id) ? captureTaskResponse(session.id, ctx.sessionOutputBuffers, ctx.taskResponseMarkers) : "";
|
|
1988
|
-
ctx.log(`session_ready for active task ${session.id}
|
|
2202
|
+
ctx.log(`session_ready for active task ${session.id} — forwarding as task_complete (stall classifier path, response: ${response.length} chars)`);
|
|
1989
2203
|
ctx.emitEvent(session.id, "task_complete", { session, response });
|
|
1990
2204
|
}
|
|
1991
2205
|
async function initializePTYManager(ctx) {
|
|
@@ -2031,7 +2245,7 @@ async function initializePTYManager(ctx) {
|
|
|
2031
2245
|
ctx.emitEvent(session.id, "task_complete", { session, response });
|
|
2032
2246
|
});
|
|
2033
2247
|
bunManager.on("tool_running", (session, info) => {
|
|
2034
|
-
ctx.log(`tool_running for ${session.id}: ${info.toolName}${info.description ? `
|
|
2248
|
+
ctx.log(`tool_running for ${session.id}: ${info.toolName}${info.description ? ` — ${info.description}` : ""}`);
|
|
2035
2249
|
ctx.emitEvent(session.id, "tool_running", { session, ...info });
|
|
2036
2250
|
});
|
|
2037
2251
|
bunManager.on("message", (message) => {
|
|
@@ -2102,7 +2316,7 @@ async function initializePTYManager(ctx) {
|
|
|
2102
2316
|
ctx.emitEvent(session.id, "task_complete", { session, response });
|
|
2103
2317
|
});
|
|
2104
2318
|
nodeManager.on("tool_running", (session, info) => {
|
|
2105
|
-
ctx.log(`tool_running for ${session.id}: ${info.toolName}${info.description ? `
|
|
2319
|
+
ctx.log(`tool_running for ${session.id}: ${info.toolName}${info.description ? ` — ${info.description}` : ""}`);
|
|
2106
2320
|
ctx.emitEvent(session.id, "tool_running", { session, ...info });
|
|
2107
2321
|
});
|
|
2108
2322
|
nodeManager.on("session_stopped", (session, reason) => {
|
|
@@ -2116,12 +2330,6 @@ async function initializePTYManager(ctx) {
|
|
|
2116
2330
|
});
|
|
2117
2331
|
return { manager: nodeManager, usingBunWorker: false };
|
|
2118
2332
|
}
|
|
2119
|
-
var _require = createRequire2(import.meta.url);
|
|
2120
|
-
var resolvedAdapterModule = "coding-agent-adapters";
|
|
2121
|
-
try {
|
|
2122
|
-
resolvedAdapterModule = _require.resolve("coding-agent-adapters");
|
|
2123
|
-
} catch {
|
|
2124
|
-
}
|
|
2125
2333
|
|
|
2126
2334
|
// src/services/pty-session-io.ts
|
|
2127
2335
|
async function sendToSession(ctx, sessionId, input) {
|
|
@@ -2176,8 +2384,7 @@ async function stopSession(ctx, sessionId, sessionMetadata, sessionWorkdirs, log
|
|
|
2176
2384
|
if (unsubscribe) {
|
|
2177
2385
|
unsubscribe();
|
|
2178
2386
|
}
|
|
2179
|
-
} catch {
|
|
2180
|
-
}
|
|
2387
|
+
} catch {}
|
|
2181
2388
|
ctx.outputUnsubscribers.delete(sessionId);
|
|
2182
2389
|
sessionMetadata.delete(sessionId);
|
|
2183
2390
|
sessionWorkdirs.delete(sessionId);
|
|
@@ -2207,7 +2414,8 @@ async function getSessionOutput(ctx, sessionId, lines) {
|
|
|
2207
2414
|
if (!buffer)
|
|
2208
2415
|
return "";
|
|
2209
2416
|
const tail = lines ?? buffer.length;
|
|
2210
|
-
return buffer.slice(-tail).join(
|
|
2417
|
+
return buffer.slice(-tail).join(`
|
|
2418
|
+
`);
|
|
2211
2419
|
}
|
|
2212
2420
|
const output = [];
|
|
2213
2421
|
for await (const line of ctx.manager.logs(sessionId, {
|
|
@@ -2215,10 +2423,26 @@ async function getSessionOutput(ctx, sessionId, lines) {
|
|
|
2215
2423
|
})) {
|
|
2216
2424
|
output.push(line);
|
|
2217
2425
|
}
|
|
2218
|
-
return output.join(
|
|
2426
|
+
return output.join(`
|
|
2427
|
+
`);
|
|
2219
2428
|
}
|
|
2220
2429
|
|
|
2221
2430
|
// src/services/pty-spawn.ts
|
|
2431
|
+
var ENV_ALLOWLIST = [
|
|
2432
|
+
"PATH",
|
|
2433
|
+
"HOME",
|
|
2434
|
+
"USER",
|
|
2435
|
+
"SHELL",
|
|
2436
|
+
"LANG",
|
|
2437
|
+
"LC_ALL",
|
|
2438
|
+
"LC_CTYPE",
|
|
2439
|
+
"TERM",
|
|
2440
|
+
"TZ",
|
|
2441
|
+
"TMPDIR",
|
|
2442
|
+
"XDG_RUNTIME_DIR",
|
|
2443
|
+
"NODE_OPTIONS",
|
|
2444
|
+
"BUN_INSTALL"
|
|
2445
|
+
];
|
|
2222
2446
|
function buildSanitizedBaseEnv() {
|
|
2223
2447
|
const env = {};
|
|
2224
2448
|
for (const key of ENV_ALLOWLIST) {
|
|
@@ -2232,7 +2456,8 @@ function setupOutputBuffer(ctx, sessionId) {
|
|
|
2232
2456
|
const buffer = [];
|
|
2233
2457
|
ctx.sessionOutputBuffers.set(sessionId, buffer);
|
|
2234
2458
|
const unsubscribe = ctx.manager.onSessionData(sessionId, (data) => {
|
|
2235
|
-
const lines = data.split(
|
|
2459
|
+
const lines = data.split(`
|
|
2460
|
+
`);
|
|
2236
2461
|
buffer.push(...lines);
|
|
2237
2462
|
while (buffer.length > (ctx.serviceConfig.maxLogLines ?? 1000)) {
|
|
2238
2463
|
buffer.shift();
|
|
@@ -2255,26 +2480,30 @@ function setupDeferredTaskDelivery(ctx, session, task, agentType) {
|
|
|
2255
2480
|
const sendTaskWithRetry = (attempt) => {
|
|
2256
2481
|
const buffer = ctx.sessionOutputBuffers.get(sid);
|
|
2257
2482
|
const baselineLength = buffer?.length ?? 0;
|
|
2258
|
-
ctx.log(`Session ${sid}
|
|
2483
|
+
ctx.log(`Session ${sid} — sending task (attempt ${attempt + 1}, ${settleMs}ms settle, baseline ${baselineLength} lines)`);
|
|
2259
2484
|
ctx.sendToSession(sid, task).catch((err) => ctx.log(`Failed to send deferred task to ${sid}: ${err}`));
|
|
2260
2485
|
if (attempt < MAX_RETRIES) {
|
|
2261
2486
|
setTimeout(() => {
|
|
2262
2487
|
const currentLength = buffer?.length ?? 0;
|
|
2263
2488
|
const newLines = currentLength - baselineLength;
|
|
2264
2489
|
if (newLines < MIN_NEW_LINES) {
|
|
2265
|
-
ctx.log(`Session ${sid}
|
|
2490
|
+
ctx.log(`Session ${sid} — task may not have been accepted (only ${newLines} new lines after ${VERIFY_DELAY_MS}ms). Retrying (attempt ${attempt + 2}/${MAX_RETRIES + 1})`);
|
|
2266
2491
|
sendTaskWithRetry(attempt + 1);
|
|
2267
2492
|
} else {
|
|
2268
|
-
ctx.log(`Session ${sid}
|
|
2493
|
+
ctx.log(`Session ${sid} — task accepted (${newLines} new lines after ${VERIFY_DELAY_MS}ms)`);
|
|
2269
2494
|
}
|
|
2270
2495
|
}, VERIFY_DELAY_MS);
|
|
2271
2496
|
}
|
|
2272
2497
|
};
|
|
2498
|
+
const READY_TIMEOUT_MS = 30000;
|
|
2273
2499
|
let taskSent = false;
|
|
2500
|
+
let readyTimeout;
|
|
2274
2501
|
const sendTask = () => {
|
|
2275
2502
|
if (taskSent)
|
|
2276
2503
|
return;
|
|
2277
2504
|
taskSent = true;
|
|
2505
|
+
if (readyTimeout)
|
|
2506
|
+
clearTimeout(readyTimeout);
|
|
2278
2507
|
setTimeout(() => sendTaskWithRetry(0), settleMs);
|
|
2279
2508
|
if (ctx.usingBunWorker) {
|
|
2280
2509
|
ctx.manager.removeListener("session_ready", onReady);
|
|
@@ -2295,6 +2524,12 @@ function setupDeferredTaskDelivery(ctx, session, task, agentType) {
|
|
|
2295
2524
|
} else {
|
|
2296
2525
|
ctx.manager.on("session_ready", onReady);
|
|
2297
2526
|
}
|
|
2527
|
+
readyTimeout = setTimeout(() => {
|
|
2528
|
+
if (!taskSent) {
|
|
2529
|
+
ctx.log(`Session ${sid} — ready event not received within ${READY_TIMEOUT_MS}ms, forcing task delivery`);
|
|
2530
|
+
sendTask();
|
|
2531
|
+
}
|
|
2532
|
+
}, READY_TIMEOUT_MS);
|
|
2298
2533
|
}
|
|
2299
2534
|
}
|
|
2300
2535
|
function buildSpawnConfig(sessionId, options, workdir) {
|
|
@@ -2329,21 +2564,6 @@ function buildSpawnConfig(sessionId, options, workdir) {
|
|
|
2329
2564
|
}
|
|
2330
2565
|
};
|
|
2331
2566
|
}
|
|
2332
|
-
var ENV_ALLOWLIST = [
|
|
2333
|
-
"PATH",
|
|
2334
|
-
"HOME",
|
|
2335
|
-
"USER",
|
|
2336
|
-
"SHELL",
|
|
2337
|
-
"LANG",
|
|
2338
|
-
"LC_ALL",
|
|
2339
|
-
"LC_CTYPE",
|
|
2340
|
-
"TERM",
|
|
2341
|
-
"TZ",
|
|
2342
|
-
"TMPDIR",
|
|
2343
|
-
"XDG_RUNTIME_DIR",
|
|
2344
|
-
"NODE_OPTIONS",
|
|
2345
|
-
"BUN_INSTALL"
|
|
2346
|
-
];
|
|
2347
2567
|
|
|
2348
2568
|
// src/services/pty-types.ts
|
|
2349
2569
|
var PI_AGENT_ALIASES = new Set([
|
|
@@ -2388,17 +2608,39 @@ var toPiCommand = (task) => {
|
|
|
2388
2608
|
|
|
2389
2609
|
// src/services/stall-classifier.ts
|
|
2390
2610
|
init_ansi_utils();
|
|
2391
|
-
import {ModelType} from "@elizaos/core";
|
|
2611
|
+
import { ModelType } from "@elizaos/core";
|
|
2392
2612
|
import {
|
|
2393
|
-
buildTaskCompletionTimeline,
|
|
2394
|
-
extractTaskCompletionTraceRecords
|
|
2613
|
+
buildTaskCompletionTimeline,
|
|
2614
|
+
extractTaskCompletionTraceRecords
|
|
2395
2615
|
} from "pty-manager";
|
|
2396
2616
|
function buildStallClassificationPrompt(agentType, sessionId, output) {
|
|
2397
|
-
return `You are Milady, an AI orchestrator managing coding agent sessions. ` + `A ${agentType} coding agent (session: ${sessionId}) appears to have stalled
|
|
2617
|
+
return `You are Milady, an AI orchestrator managing coding agent sessions. ` + `A ${agentType} coding agent (session: ${sessionId}) appears to have stalled — ` + `it has stopped producing output while in a busy state.
|
|
2618
|
+
|
|
2619
|
+
` + `Here is the recent terminal output:
|
|
2620
|
+
` + `---
|
|
2621
|
+
${output.slice(-1500)}
|
|
2622
|
+
---
|
|
2623
|
+
|
|
2624
|
+
` + `Classify what's happening. Read the output carefully and choose the MOST specific match:
|
|
2625
|
+
|
|
2626
|
+
` + `1. "task_complete" — The agent FINISHED its task and returned to its idle prompt. ` + `Strong indicators: a summary of completed work ("Done", "All done", "Here's what was completed"), ` + `timing info ("Baked for", "Churned for", "Crunched for", "Cooked for", "Worked for"), ` + `or the agent's main prompt symbol (❯) appearing AFTER completion output. ` + `If the output contains evidence of completed work followed by an idle prompt, this is ALWAYS task_complete, ` + `even though the agent is technically "waiting" — it is waiting for a NEW task, not asking a question.
|
|
2627
|
+
|
|
2628
|
+
` + `2. "waiting_for_input" — The agent is MID-TASK and blocked on a specific question or permission prompt. ` + `The agent has NOT finished its work — it needs a response to continue. ` + `Examples: Y/n confirmation, file permission dialogs, "Do you want to proceed?", ` + `tool approval prompts, or interactive menus. ` + `This is NOT the same as the agent sitting at its idle prompt after finishing work.
|
|
2629
|
+
|
|
2630
|
+
` + `3. "still_working" — The agent is actively processing (API call, compilation, thinking, etc.) ` + `and has not produced final output yet. No prompt or completion summary visible.
|
|
2631
|
+
|
|
2632
|
+
` + `4. "error" — The agent hit an error state (crash, unrecoverable error, stack trace).
|
|
2633
|
+
|
|
2634
|
+
` + `5. "tool_running" — The agent is using an external tool (browser automation, ` + `MCP tool, etc.). Indicators: "Claude in Chrome", "javascript_tool", ` + `"computer_tool", "screenshot", "navigate", tool execution output. ` + `The agent is actively working but the terminal may be quiet.
|
|
2398
2635
|
|
|
2399
|
-
` + `
|
|
2636
|
+
` + `IMPORTANT: If you see BOTH completed work output AND an idle prompt (❯), choose "task_complete". ` + `Only choose "waiting_for_input" if the agent is clearly asking a question mid-task.
|
|
2400
2637
|
|
|
2401
|
-
` + `
|
|
2638
|
+
` + `If "waiting_for_input", also provide:
|
|
2639
|
+
` + `- "prompt": the text of what it's asking
|
|
2640
|
+
` + `- "suggestedResponse": what to type/send. Use "keys:enter" for TUI menu confirmation, ` + `"keys:down,enter" to select a non-default option, or plain text like "y" for text prompts.
|
|
2641
|
+
|
|
2642
|
+
` + `Respond with ONLY a JSON object:
|
|
2643
|
+
` + `{"state": "...", "prompt": "...", "suggestedResponse": "..."}`;
|
|
2402
2644
|
}
|
|
2403
2645
|
async function writeStallSnapshot(sessionId, agentType, recentOutput, effectiveOutput, buffers, traceEntries, log) {
|
|
2404
2646
|
try {
|
|
@@ -2408,7 +2650,8 @@ async function writeStallSnapshot(sessionId, agentType, recentOutput, effectiveO
|
|
|
2408
2650
|
const snapshotDir = path.join(os.homedir(), ".milady", "debug");
|
|
2409
2651
|
fs.mkdirSync(snapshotDir, { recursive: true });
|
|
2410
2652
|
const ourBuffer = buffers.get(sessionId);
|
|
2411
|
-
const ourTail = ourBuffer ? ourBuffer.slice(-100).join(
|
|
2653
|
+
const ourTail = ourBuffer ? ourBuffer.slice(-100).join(`
|
|
2654
|
+
`) : "(no buffer)";
|
|
2412
2655
|
let traceTimeline = "(no trace entries)";
|
|
2413
2656
|
try {
|
|
2414
2657
|
const records = extractTaskCompletionTraceRecords(traceEntries);
|
|
@@ -2431,14 +2674,15 @@ async function writeStallSnapshot(sessionId, agentType, recentOutput, effectiveO
|
|
|
2431
2674
|
traceTimeline,
|
|
2432
2675
|
``,
|
|
2433
2676
|
`--- raw trace entries (last 20 of ${traceEntries.length}) ---`,
|
|
2434
|
-
traceEntries.slice(-20).join(
|
|
2677
|
+
traceEntries.slice(-20).join(`
|
|
2678
|
+
`),
|
|
2435
2679
|
``
|
|
2436
|
-
].join(
|
|
2680
|
+
].join(`
|
|
2681
|
+
`);
|
|
2437
2682
|
const snapshotPath = path.join(snapshotDir, `stall-snapshot-${sessionId}.txt`);
|
|
2438
2683
|
fs.writeFileSync(snapshotPath, snapshot);
|
|
2439
|
-
log(`Stall snapshot
|
|
2440
|
-
} catch (_) {
|
|
2441
|
-
}
|
|
2684
|
+
log(`Stall snapshot → ${snapshotPath}`);
|
|
2685
|
+
} catch (_) {}
|
|
2442
2686
|
}
|
|
2443
2687
|
async function classifyStallOutput(ctx) {
|
|
2444
2688
|
const {
|
|
@@ -2457,7 +2701,8 @@ async function classifyStallOutput(ctx) {
|
|
|
2457
2701
|
if (!recentOutput || recentOutput.trim().length < 200) {
|
|
2458
2702
|
const ourBuffer = buffers.get(sessionId);
|
|
2459
2703
|
if (ourBuffer && ourBuffer.length > 0) {
|
|
2460
|
-
const rawTail = ourBuffer.slice(-100).join(
|
|
2704
|
+
const rawTail = ourBuffer.slice(-100).join(`
|
|
2705
|
+
`);
|
|
2461
2706
|
const stripped = stripAnsi(rawTail);
|
|
2462
2707
|
if (stripped.length > effectiveOutput.length) {
|
|
2463
2708
|
effectiveOutput = stripped;
|
|
@@ -2497,7 +2742,7 @@ async function classifyStallOutput(ctx) {
|
|
|
2497
2742
|
prompt: parsed.prompt,
|
|
2498
2743
|
suggestedResponse: parsed.suggestedResponse
|
|
2499
2744
|
};
|
|
2500
|
-
log(`Stall classification for ${sessionId}: ${classification.state}${classification.suggestedResponse ? `
|
|
2745
|
+
log(`Stall classification for ${sessionId}: ${classification.state}${classification.suggestedResponse ? ` → "${classification.suggestedResponse}"` : ""}`);
|
|
2501
2746
|
if (classification.state === "task_complete") {
|
|
2502
2747
|
const session = manager?.get(sessionId);
|
|
2503
2748
|
const durationMs = session?.startedAt ? Date.now() - new Date(session.startedAt).getTime() : 0;
|
|
@@ -2513,12 +2758,14 @@ async function classifyStallOutput(ctx) {
|
|
|
2513
2758
|
// src/services/swarm-coordinator.ts
|
|
2514
2759
|
init_ansi_utils();
|
|
2515
2760
|
init_swarm_decision_loop();
|
|
2516
|
-
import {logger} from "@elizaos/core";
|
|
2761
|
+
import { logger } from "@elizaos/core";
|
|
2517
2762
|
|
|
2518
2763
|
// src/services/swarm-idle-watchdog.ts
|
|
2519
2764
|
init_ansi_utils();
|
|
2520
2765
|
init_swarm_decision_loop();
|
|
2521
|
-
import {ModelType as ModelType4} from "@elizaos/core";
|
|
2766
|
+
import { ModelType as ModelType4 } from "@elizaos/core";
|
|
2767
|
+
var IDLE_THRESHOLD_MS = 5 * 60 * 1000;
|
|
2768
|
+
var MAX_IDLE_CHECKS = 4;
|
|
2522
2769
|
async function scanIdleSessions(ctx) {
|
|
2523
2770
|
const now = Date.now();
|
|
2524
2771
|
for (const taskCtx of ctx.tasks.values()) {
|
|
@@ -2527,7 +2774,7 @@ async function scanIdleSessions(ctx) {
|
|
|
2527
2774
|
if (ctx.ptyService) {
|
|
2528
2775
|
const session = ctx.ptyService.getSession(taskCtx.sessionId);
|
|
2529
2776
|
if (!session) {
|
|
2530
|
-
ctx.log(`Idle watchdog: "${taskCtx.label}"
|
|
2777
|
+
ctx.log(`Idle watchdog: "${taskCtx.label}" — PTY session no longer exists, marking as stopped`);
|
|
2531
2778
|
taskCtx.status = "stopped";
|
|
2532
2779
|
taskCtx.decisions.push({
|
|
2533
2780
|
timestamp: now,
|
|
@@ -2542,7 +2789,7 @@ async function scanIdleSessions(ctx) {
|
|
|
2542
2789
|
timestamp: now,
|
|
2543
2790
|
data: { reason: "pty_session_gone" }
|
|
2544
2791
|
});
|
|
2545
|
-
ctx.sendChatMessage(`[${taskCtx.label}] Session lost
|
|
2792
|
+
ctx.sendChatMessage(`[${taskCtx.label}] Session lost — the agent process is no longer running (likely killed during a restart).`, "coding-agent");
|
|
2546
2793
|
checkAllTasksComplete(ctx);
|
|
2547
2794
|
continue;
|
|
2548
2795
|
}
|
|
@@ -2561,11 +2808,10 @@ async function scanIdleSessions(ctx) {
|
|
|
2561
2808
|
if (currentOutput !== lastSeen) {
|
|
2562
2809
|
taskCtx.lastActivityAt = now;
|
|
2563
2810
|
taskCtx.idleCheckCount = 0;
|
|
2564
|
-
ctx.log(`Idle watchdog: "${taskCtx.label}" has fresh PTY output
|
|
2811
|
+
ctx.log(`Idle watchdog: "${taskCtx.label}" has fresh PTY output — not idle`);
|
|
2565
2812
|
continue;
|
|
2566
2813
|
}
|
|
2567
|
-
} catch {
|
|
2568
|
-
}
|
|
2814
|
+
} catch {}
|
|
2569
2815
|
}
|
|
2570
2816
|
taskCtx.idleCheckCount++;
|
|
2571
2817
|
const idleMinutes = Math.round(idleMs / 60000);
|
|
@@ -2590,7 +2836,7 @@ async function scanIdleSessions(ctx) {
|
|
|
2590
2836
|
idleCheckCount: taskCtx.idleCheckCount
|
|
2591
2837
|
}
|
|
2592
2838
|
});
|
|
2593
|
-
ctx.sendChatMessage(`[${taskCtx.label}] Session stopped
|
|
2839
|
+
ctx.sendChatMessage(`[${taskCtx.label}] Session stopped — idle for ${idleMinutes} minutes with no progress.`, "coding-agent");
|
|
2594
2840
|
if (ctx.ptyService) {
|
|
2595
2841
|
try {
|
|
2596
2842
|
await ctx.ptyService.stopSession(taskCtx.sessionId, true);
|
|
@@ -2649,8 +2895,8 @@ async function handleIdleCheck(ctx, taskCtx, idleMinutes) {
|
|
|
2649
2895
|
ctx.log(`Idle check LLM call failed: ${err}`);
|
|
2650
2896
|
}
|
|
2651
2897
|
if (!decision) {
|
|
2652
|
-
ctx.log(`Idle check for "${taskCtx.label}": LLM returned invalid response
|
|
2653
|
-
ctx.sendChatMessage(`[${taskCtx.label}] Session idle for ${idleMinutes}m
|
|
2898
|
+
ctx.log(`Idle check for "${taskCtx.label}": LLM returned invalid response — escalating`);
|
|
2899
|
+
ctx.sendChatMessage(`[${taskCtx.label}] Session idle for ${idleMinutes}m — couldn't determine status. Needs your attention.`, "coding-agent");
|
|
2654
2900
|
return;
|
|
2655
2901
|
}
|
|
2656
2902
|
taskCtx.decisions.push({
|
|
@@ -2672,22 +2918,19 @@ async function handleIdleCheck(ctx, taskCtx, idleMinutes) {
|
|
|
2672
2918
|
reasoning: decision.reasoning
|
|
2673
2919
|
}
|
|
2674
2920
|
});
|
|
2675
|
-
if (decision.action === "complete") {
|
|
2676
|
-
} else if (decision.action === "respond") {
|
|
2921
|
+
if (decision.action === "complete") {} else if (decision.action === "respond") {
|
|
2677
2922
|
const actionDesc = decision.useKeys ? `Sent keys: ${decision.keys?.join(", ")}` : `Nudged: ${decision.response ?? ""}`;
|
|
2678
|
-
ctx.sendChatMessage(`[${taskCtx.label}] Idle for ${idleMinutes}m
|
|
2923
|
+
ctx.sendChatMessage(`[${taskCtx.label}] Idle for ${idleMinutes}m — ${actionDesc}`, "coding-agent");
|
|
2679
2924
|
} else if (decision.action === "escalate") {
|
|
2680
|
-
ctx.sendChatMessage(`[${taskCtx.label}] Idle for ${idleMinutes}m
|
|
2925
|
+
ctx.sendChatMessage(`[${taskCtx.label}] Idle for ${idleMinutes}m — needs your attention: ${decision.reasoning}`, "coding-agent");
|
|
2681
2926
|
} else if (decision.action === "ignore") {
|
|
2682
|
-
ctx.log(`Idle check for "${taskCtx.label}": LLM says still working
|
|
2927
|
+
ctx.log(`Idle check for "${taskCtx.label}": LLM says still working — ${decision.reasoning}`);
|
|
2683
2928
|
}
|
|
2684
2929
|
await executeDecision(ctx, sessionId, decision);
|
|
2685
2930
|
} finally {
|
|
2686
2931
|
ctx.inFlightDecisions.delete(sessionId);
|
|
2687
2932
|
}
|
|
2688
2933
|
}
|
|
2689
|
-
var IDLE_THRESHOLD_MS = 5 * 60 * 1000;
|
|
2690
|
-
var MAX_IDLE_CHECKS = 4;
|
|
2691
2934
|
|
|
2692
2935
|
// src/services/swarm-coordinator.ts
|
|
2693
2936
|
var UNREGISTERED_BUFFER_MS = 2000;
|
|
@@ -2727,7 +2970,7 @@ class SwarmCoordinator {
|
|
|
2727
2970
|
}
|
|
2728
2971
|
setAgentDecisionCallback(cb) {
|
|
2729
2972
|
this.agentDecisionCb = cb;
|
|
2730
|
-
this.log("Agent decision callback wired
|
|
2973
|
+
this.log("Agent decision callback wired — events will route through Milaidy");
|
|
2731
2974
|
}
|
|
2732
2975
|
getAgentDecisionCallback() {
|
|
2733
2976
|
return this.agentDecisionCb;
|
|
@@ -2790,7 +3033,7 @@ class SwarmCoordinator {
|
|
|
2790
3033
|
if (this._paused)
|
|
2791
3034
|
return;
|
|
2792
3035
|
this._paused = true;
|
|
2793
|
-
this.log("Coordinator paused
|
|
3036
|
+
this.log("Coordinator paused — buffering LLM decisions until user message is processed");
|
|
2794
3037
|
this.broadcast({ type: "coordinator_paused", sessionId: "", timestamp: Date.now(), data: {} });
|
|
2795
3038
|
this.pauseTimeout = setTimeout(() => {
|
|
2796
3039
|
if (this._paused) {
|
|
@@ -2807,7 +3050,7 @@ class SwarmCoordinator {
|
|
|
2807
3050
|
clearTimeout(this.pauseTimeout);
|
|
2808
3051
|
this.pauseTimeout = null;
|
|
2809
3052
|
}
|
|
2810
|
-
this.log(`Coordinator resumed
|
|
3053
|
+
this.log(`Coordinator resumed — replaying ${this.pauseBuffer.length} buffered events`);
|
|
2811
3054
|
this.broadcast({ type: "coordinator_resumed", sessionId: "", timestamp: Date.now(), data: {} });
|
|
2812
3055
|
const buffered = [...this.pauseBuffer];
|
|
2813
3056
|
this.pauseBuffer = [];
|
|
@@ -2902,9 +3145,10 @@ class SwarmCoordinator {
|
|
|
2902
3145
|
}
|
|
2903
3146
|
writeSseEvent(res, event) {
|
|
2904
3147
|
try {
|
|
2905
|
-
res.write(`data: ${JSON.stringify(event)}
|
|
2906
|
-
|
|
2907
|
-
|
|
3148
|
+
res.write(`data: ${JSON.stringify(event)}
|
|
3149
|
+
|
|
3150
|
+
`);
|
|
3151
|
+
} catch {}
|
|
2908
3152
|
}
|
|
2909
3153
|
async handleSessionEvent(sessionId, event, data) {
|
|
2910
3154
|
const taskCtx = this.tasks.get(sessionId);
|
|
@@ -2923,8 +3167,7 @@ class SwarmCoordinator {
|
|
|
2923
3167
|
if (ctx) {
|
|
2924
3168
|
this.unregisteredBuffer.delete(sessionId);
|
|
2925
3169
|
for (const entry of stillBuffered) {
|
|
2926
|
-
this.handleSessionEvent(sessionId, entry.event, entry.data).catch(() => {
|
|
2927
|
-
});
|
|
3170
|
+
this.handleSessionEvent(sessionId, entry.event, entry.data).catch(() => {});
|
|
2928
3171
|
}
|
|
2929
3172
|
} else {
|
|
2930
3173
|
this.unregisteredBuffer.delete(sessionId);
|
|
@@ -3030,10 +3273,9 @@ class SwarmCoordinator {
|
|
|
3030
3273
|
if (devUrl) {
|
|
3031
3274
|
urlSuffix = ` Dev server running at ${devUrl}`;
|
|
3032
3275
|
}
|
|
3033
|
-
} catch {
|
|
3034
|
-
}
|
|
3276
|
+
} catch {}
|
|
3035
3277
|
}
|
|
3036
|
-
this.sendChatMessage(`[${taskCtx.label}] Running ${toolDesc}.${urlSuffix} The agent is working outside the terminal
|
|
3278
|
+
this.sendChatMessage(`[${taskCtx.label}] Running ${toolDesc}.${urlSuffix} The agent is working outside the terminal — I'll let it finish.`, "coding-agent");
|
|
3037
3279
|
}
|
|
3038
3280
|
break;
|
|
3039
3281
|
}
|
|
@@ -3172,14 +3414,21 @@ class PTYService {
|
|
|
3172
3414
|
const config = runtime.getSetting("PTY_SERVICE_CONFIG");
|
|
3173
3415
|
const service = new PTYService(runtime, config ?? {});
|
|
3174
3416
|
await service.initialize();
|
|
3175
|
-
|
|
3176
|
-
|
|
3177
|
-
|
|
3178
|
-
service.coordinator =
|
|
3179
|
-
|
|
3180
|
-
|
|
3181
|
-
|
|
3182
|
-
|
|
3417
|
+
const servicesMap = runtime.services;
|
|
3418
|
+
const existing = servicesMap?.get?.("SWARM_COORDINATOR");
|
|
3419
|
+
if (existing && existing.length > 0) {
|
|
3420
|
+
service.coordinator = existing[0];
|
|
3421
|
+
logger2.info("[PTYService] SwarmCoordinator already registered, skipping duplicate start");
|
|
3422
|
+
} else {
|
|
3423
|
+
try {
|
|
3424
|
+
const coordinator = new SwarmCoordinator(runtime);
|
|
3425
|
+
coordinator.start(service);
|
|
3426
|
+
service.coordinator = coordinator;
|
|
3427
|
+
servicesMap?.set?.("SWARM_COORDINATOR", [coordinator]);
|
|
3428
|
+
logger2.info("[PTYService] SwarmCoordinator wired and started");
|
|
3429
|
+
} catch (err) {
|
|
3430
|
+
logger2.error(`[PTYService] Failed to wire SwarmCoordinator: ${err}`);
|
|
3431
|
+
}
|
|
3183
3432
|
}
|
|
3184
3433
|
return service;
|
|
3185
3434
|
}
|
|
@@ -3305,8 +3554,7 @@ class PTYService {
|
|
|
3305
3554
|
let settings = {};
|
|
3306
3555
|
try {
|
|
3307
3556
|
settings = JSON.parse(await readFile(settingsPath, "utf-8"));
|
|
3308
|
-
} catch {
|
|
3309
|
-
}
|
|
3557
|
+
} catch {}
|
|
3310
3558
|
const permissions = settings.permissions ?? {};
|
|
3311
3559
|
permissions.allowedDirectories = [workdir];
|
|
3312
3560
|
settings.permissions = permissions;
|
|
@@ -3678,7 +3926,9 @@ var spawnAgentAction = {
|
|
|
3678
3926
|
if (preflight && !preflight.installed) {
|
|
3679
3927
|
if (callback) {
|
|
3680
3928
|
await callback({
|
|
3681
|
-
text: `${preflight.adapter} CLI is not installed
|
|
3929
|
+
text: `${preflight.adapter} CLI is not installed.
|
|
3930
|
+
` + `Install with: ${preflight.installCommand}
|
|
3931
|
+
` + `Docs: ${preflight.docsUrl}`
|
|
3682
3932
|
});
|
|
3683
3933
|
}
|
|
3684
3934
|
return { success: false, error: "AGENT_NOT_INSTALLED" };
|
|
@@ -3804,16 +4054,16 @@ var spawnAgentAction = {
|
|
|
3804
4054
|
|
|
3805
4055
|
// src/actions/coding-task-handlers.ts
|
|
3806
4056
|
import {
|
|
3807
|
-
logger as logger5
|
|
4057
|
+
logger as logger5
|
|
3808
4058
|
} from "@elizaos/core";
|
|
3809
4059
|
|
|
3810
4060
|
// src/actions/coding-task-helpers.ts
|
|
3811
|
-
import {randomUUID} from "node:crypto";
|
|
4061
|
+
import { randomUUID } from "node:crypto";
|
|
3812
4062
|
import * as fs from "node:fs";
|
|
3813
4063
|
import * as os2 from "node:os";
|
|
3814
4064
|
import * as path3 from "node:path";
|
|
3815
4065
|
import {
|
|
3816
|
-
logger as logger4
|
|
4066
|
+
logger as logger4
|
|
3817
4067
|
} from "@elizaos/core";
|
|
3818
4068
|
function createScratchDir() {
|
|
3819
4069
|
const baseDir = path3.join(os2.homedir(), ".milady", "workspaces");
|
|
@@ -3852,7 +4102,9 @@ function registerSessionEvents(ptyService, runtime, sessionId, label, scratchDir
|
|
|
3852
4102
|
const response = data.response ?? "";
|
|
3853
4103
|
const preview = response.length > 500 ? `${response.slice(0, 500)}...` : response;
|
|
3854
4104
|
callback({
|
|
3855
|
-
text: preview ? `Agent "${label}" completed the task
|
|
4105
|
+
text: preview ? `Agent "${label}" completed the task.
|
|
4106
|
+
|
|
4107
|
+
${preview}` : `Agent "${label}" completed the task.`
|
|
3856
4108
|
});
|
|
3857
4109
|
}
|
|
3858
4110
|
ptyService.stopSession(sessionId, true).catch((err) => {
|
|
@@ -3877,6 +4129,7 @@ function registerSessionEvents(ptyService, runtime, sessionId, label, scratchDir
|
|
|
3877
4129
|
}
|
|
3878
4130
|
|
|
3879
4131
|
// src/actions/coding-task-handlers.ts
|
|
4132
|
+
var MAX_CONCURRENT_AGENTS = 8;
|
|
3880
4133
|
async function handleMultiAgent(ctx, agentsParam) {
|
|
3881
4134
|
const {
|
|
3882
4135
|
runtime,
|
|
@@ -4057,7 +4310,8 @@ async function handleMultiAgent(ctx, agentsParam) {
|
|
|
4057
4310
|
`Launched ${succeeded.length}/${agentSpecs.length} agents${repo ? ` on ${repo}` : ""}:`,
|
|
4058
4311
|
...succeeded.map((r) => ` - "${r.label}" (${r.agentType}) [session: ${r.sessionId}]`),
|
|
4059
4312
|
...failed.length > 0 ? [`Failed: ${failed.map((r) => `"${r.label}": ${r.error}`).join(", ")}`] : []
|
|
4060
|
-
].join(
|
|
4313
|
+
].join(`
|
|
4314
|
+
`);
|
|
4061
4315
|
if (callback) {
|
|
4062
4316
|
await callback({ text: summary });
|
|
4063
4317
|
}
|
|
@@ -4138,7 +4392,9 @@ async function handleSingleAgent(ctx, task) {
|
|
|
4138
4392
|
logger5.warn(`[START_CODING_TASK] ${preflight.adapter} CLI not installed`);
|
|
4139
4393
|
if (callback) {
|
|
4140
4394
|
await callback({
|
|
4141
|
-
text: `${preflight.adapter} CLI is not installed
|
|
4395
|
+
text: `${preflight.adapter} CLI is not installed.
|
|
4396
|
+
Install with: ${preflight.installCommand}
|
|
4397
|
+
Docs: ${preflight.docsUrl}`
|
|
4142
4398
|
});
|
|
4143
4399
|
}
|
|
4144
4400
|
return { success: false, error: "AGENT_NOT_INSTALLED" };
|
|
@@ -4191,7 +4447,8 @@ async function handleSingleAgent(ctx, task) {
|
|
|
4191
4447
|
}
|
|
4192
4448
|
const summary = repo ? `Cloned ${repo} and started ${displayType} agent as "${label}"${task ? ` with task: "${task}"` : ""}` : `Started ${displayType} agent as "${label}" in scratch workspace${task ? ` with task: "${task}"` : ""}`;
|
|
4193
4449
|
if (callback) {
|
|
4194
|
-
await callback({ text: `${summary}
|
|
4450
|
+
await callback({ text: `${summary}
|
|
4451
|
+
Session ID: ${session.id}` });
|
|
4195
4452
|
}
|
|
4196
4453
|
return {
|
|
4197
4454
|
success: true,
|
|
@@ -4217,7 +4474,6 @@ async function handleSingleAgent(ctx, task) {
|
|
|
4217
4474
|
return { success: false, error: errorMessage };
|
|
4218
4475
|
}
|
|
4219
4476
|
}
|
|
4220
|
-
var MAX_CONCURRENT_AGENTS = 8;
|
|
4221
4477
|
|
|
4222
4478
|
// src/actions/start-coding-task.ts
|
|
4223
4479
|
var startCodingTaskAction = {
|
|
@@ -4229,7 +4485,7 @@ var startCodingTaskAction = {
|
|
|
4229
4485
|
"SPAWN_AND_PROVISION",
|
|
4230
4486
|
"CODE_THIS"
|
|
4231
4487
|
],
|
|
4232
|
-
description: "Start a coding task: optionally clone a repo, then spawn a coding agent (Claude Code, Codex, Gemini, Aider, Pi) " + "to work on it. If no repo is provided, the agent runs in a safe scratch directory. " + "Use this whenever the user asks to work on code, research something with an agent, or run any agent task. " + "IMPORTANT: If the user references a repository from conversation history (e.g. 'in the same repo', " + "'on that project', 'add a feature to it'), you MUST include the repo URL in the `repo` parameter. " + "If the task involves code changes to a real project but you don't know the repo URL, ASK the user for it " + "before calling this action
|
|
4488
|
+
description: "Start a coding task: optionally clone a repo, then spawn a coding agent (Claude Code, Codex, Gemini, Aider, Pi) " + "to work on it. If no repo is provided, the agent runs in a safe scratch directory. " + "Use this whenever the user asks to work on code, research something with an agent, or run any agent task. " + "IMPORTANT: If the user references a repository from conversation history (e.g. 'in the same repo', " + "'on that project', 'add a feature to it'), you MUST include the repo URL in the `repo` parameter. " + "If the task involves code changes to a real project but you don't know the repo URL, ASK the user for it " + "before calling this action — do not default to a scratch directory for real project work.",
|
|
4233
4489
|
examples: [
|
|
4234
4490
|
[
|
|
4235
4491
|
{
|
|
@@ -4389,7 +4645,7 @@ var startCodingTaskAction = {
|
|
|
4389
4645
|
|
|
4390
4646
|
// src/actions/stop-agent.ts
|
|
4391
4647
|
import {
|
|
4392
|
-
logger as logger6
|
|
4648
|
+
logger as logger6
|
|
4393
4649
|
} from "@elizaos/core";
|
|
4394
4650
|
var stopAgentAction = {
|
|
4395
4651
|
name: "STOP_CODING_AGENT",
|
|
@@ -4552,15 +4808,6 @@ var stopAgentAction = {
|
|
|
4552
4808
|
};
|
|
4553
4809
|
|
|
4554
4810
|
// src/providers/action-examples.ts
|
|
4555
|
-
function formatExample(ex) {
|
|
4556
|
-
const actionTags = ex.actions.map((a) => ` <action>${a}</action>`).join("\n");
|
|
4557
|
-
const paramBlocks = Object.entries(ex.params ?? {}).map(([actionName, params]) => {
|
|
4558
|
-
const inner = Object.entries(params).map(([k, v]) => ` <${k}>${v}</${k}>`).join("\n");
|
|
4559
|
-
return ` <${actionName}>\n${inner}\n </${actionName}>`;
|
|
4560
|
-
}).join("\n");
|
|
4561
|
-
const paramsSection = paramBlocks ? `\n<params>\n${paramBlocks}\n</params>` : "";
|
|
4562
|
-
return `User: ${ex.user}\nAssistant:\n<actions>\n${actionTags}\n</actions>${paramsSection}`;
|
|
4563
|
-
}
|
|
4564
4811
|
var CODING_AGENT_EXAMPLES = [
|
|
4565
4812
|
{
|
|
4566
4813
|
user: "Can you set up a workspace for https://github.com/acme/my-app and have Claude fix the login bug?",
|
|
@@ -4569,7 +4816,7 @@ var CODING_AGENT_EXAMPLES = [
|
|
|
4569
4816
|
START_CODING_TASK: {
|
|
4570
4817
|
repo: "https://github.com/acme/my-app",
|
|
4571
4818
|
agentType: "claude",
|
|
4572
|
-
task: "Fix the login bug in src/auth.ts
|
|
4819
|
+
task: "Fix the login bug in src/auth.ts — users are getting 401 errors after token refresh"
|
|
4573
4820
|
}
|
|
4574
4821
|
}
|
|
4575
4822
|
},
|
|
@@ -4622,7 +4869,28 @@ var CODING_AGENT_EXAMPLES = [
|
|
|
4622
4869
|
}
|
|
4623
4870
|
}
|
|
4624
4871
|
];
|
|
4625
|
-
|
|
4872
|
+
function formatExample(ex) {
|
|
4873
|
+
const actionTags = ex.actions.map((a) => ` <action>${a}</action>`).join(`
|
|
4874
|
+
`);
|
|
4875
|
+
const paramBlocks = Object.entries(ex.params ?? {}).map(([actionName, params]) => {
|
|
4876
|
+
const inner = Object.entries(params).map(([k, v]) => ` <${k}>${v}</${k}>`).join(`
|
|
4877
|
+
`);
|
|
4878
|
+
return ` <${actionName}>
|
|
4879
|
+
${inner}
|
|
4880
|
+
</${actionName}>`;
|
|
4881
|
+
}).join(`
|
|
4882
|
+
`);
|
|
4883
|
+
const paramsSection = paramBlocks ? `
|
|
4884
|
+
<params>
|
|
4885
|
+
${paramBlocks}
|
|
4886
|
+
</params>` : "";
|
|
4887
|
+
return `User: ${ex.user}
|
|
4888
|
+
Assistant:
|
|
4889
|
+
<actions>
|
|
4890
|
+
${actionTags}
|
|
4891
|
+
</actions>${paramsSection}`;
|
|
4892
|
+
}
|
|
4893
|
+
var MULTI_AGENT_EXAMPLE = `User: Spin up 3 agents on https://github.com/acme/app — one to fix auth, one to write tests, one to update docs
|
|
4626
4894
|
Assistant:
|
|
4627
4895
|
<actions>
|
|
4628
4896
|
<action>REPLY</action>
|
|
@@ -4631,7 +4899,7 @@ Assistant:
|
|
|
4631
4899
|
<params>
|
|
4632
4900
|
<START_CODING_TASK>
|
|
4633
4901
|
<repo>https://github.com/acme/app</repo>
|
|
4634
|
-
<agents>Fix the authentication bug in src/auth.ts
|
|
4902
|
+
<agents>Fix the authentication bug in src/auth.ts — users get 401 after token refresh. Your unique identifier is "alpha". | Write comprehensive unit tests for the auth module in src/auth.ts. Your unique identifier is "beta". | Update the API documentation in docs/ to reflect the new auth flow. Your unique identifier is "gamma".</agents>
|
|
4635
4903
|
</START_CODING_TASK>
|
|
4636
4904
|
</params>`;
|
|
4637
4905
|
var codingAgentExamplesProvider = {
|
|
@@ -4639,12 +4907,14 @@ var codingAgentExamplesProvider = {
|
|
|
4639
4907
|
description: "Structured examples showing how to use coding agent actions with parameters",
|
|
4640
4908
|
position: -1,
|
|
4641
4909
|
get: async (_runtime, _message, _state) => {
|
|
4642
|
-
const examples = CODING_AGENT_EXAMPLES.map(formatExample).join(
|
|
4910
|
+
const examples = CODING_AGENT_EXAMPLES.map(formatExample).join(`
|
|
4911
|
+
|
|
4912
|
+
`);
|
|
4643
4913
|
const text = [
|
|
4644
4914
|
"# Coding Agent Action Call Examples",
|
|
4645
4915
|
"When the user asks you to work on code, clone repos, spawn agents, or run agent tasks,",
|
|
4646
4916
|
"you MUST select the appropriate actions and include parameters. Do NOT just describe",
|
|
4647
|
-
"what you would do
|
|
4917
|
+
"what you would do — actually select the actions.",
|
|
4648
4918
|
"",
|
|
4649
4919
|
"IMPORTANT: Use START_CODING_TASK to launch coding agents. It handles workspace setup",
|
|
4650
4920
|
"automatically. If a repo URL is provided, it clones it first. If no repo, the agent",
|
|
@@ -4662,7 +4932,8 @@ var codingAgentExamplesProvider = {
|
|
|
4662
4932
|
"and unique identifiers so their work is clearly differentiated.",
|
|
4663
4933
|
"",
|
|
4664
4934
|
MULTI_AGENT_EXAMPLE
|
|
4665
|
-
].join(
|
|
4935
|
+
].join(`
|
|
4936
|
+
`);
|
|
4666
4937
|
return {
|
|
4667
4938
|
data: { codingAgentExamples: CODING_AGENT_EXAMPLES },
|
|
4668
4939
|
values: { codingAgentExamples: text },
|
|
@@ -4695,7 +4966,7 @@ function formatWorkspaceLine(ws, sessions) {
|
|
|
4695
4966
|
const label = ws.label || ws.id.slice(0, 8);
|
|
4696
4967
|
const agents = sessions.filter((s) => s.workdir === ws.path);
|
|
4697
4968
|
const agentSummary = agents.length > 0 ? agents.map((a) => `${a.agentType}:${formatStatus(a.status)}`).join(", ") : "no agents";
|
|
4698
|
-
return ` - "${label}"
|
|
4969
|
+
return ` - "${label}" → ${ws.repo} (branch: ${ws.branch}, ${agentSummary})`;
|
|
4699
4970
|
}
|
|
4700
4971
|
var activeWorkspaceContextProvider = {
|
|
4701
4972
|
name: "ACTIVE_WORKSPACE_CONTEXT",
|
|
@@ -4724,7 +4995,8 @@ var activeWorkspaceContextProvider = {
|
|
|
4724
4995
|
"# Active Workspaces & Agents",
|
|
4725
4996
|
"No active workspaces or coding agent sessions.",
|
|
4726
4997
|
"Use START_CODING_TASK to launch a new coding agent."
|
|
4727
|
-
].join(
|
|
4998
|
+
].join(`
|
|
4999
|
+
`);
|
|
4728
5000
|
return {
|
|
4729
5001
|
data: { activeWorkspaces: [], activeSessions: [] },
|
|
4730
5002
|
values: { activeWorkspaceContext: text2 },
|
|
@@ -4754,9 +5026,9 @@ var activeWorkspaceContextProvider = {
|
|
|
4754
5026
|
const supervisionLevel = coordinator.getSupervisionLevel();
|
|
4755
5027
|
if (pending.length > 0) {
|
|
4756
5028
|
lines.push("");
|
|
4757
|
-
lines.push(`## Pending Confirmations (${pending.length})
|
|
5029
|
+
lines.push(`## Pending Confirmations (${pending.length}) — supervision: ${supervisionLevel}`);
|
|
4758
5030
|
for (const p of pending) {
|
|
4759
|
-
lines.push(` - "${p.taskContext.label}" blocked: "${p.promptText}"
|
|
5031
|
+
lines.push(` - "${p.taskContext.label}" blocked: "${p.promptText}" → suggested: ${p.llmDecision.action}`);
|
|
4760
5032
|
}
|
|
4761
5033
|
} else if (supervisionLevel !== "autonomous") {
|
|
4762
5034
|
lines.push("");
|
|
@@ -4767,7 +5039,8 @@ var activeWorkspaceContextProvider = {
|
|
|
4767
5039
|
lines.push("");
|
|
4768
5040
|
lines.push("You can interact with agents using SEND_TO_CODING_AGENT (pass sessionId), " + "stop them with STOP_CODING_AGENT, or finalize their work with FINALIZE_WORKSPACE.");
|
|
4769
5041
|
}
|
|
4770
|
-
const text = lines.join(
|
|
5042
|
+
const text = lines.join(`
|
|
5043
|
+
`);
|
|
4771
5044
|
return {
|
|
4772
5045
|
data: {
|
|
4773
5046
|
activeWorkspaces: workspaces.map((ws) => ({
|
|
@@ -4795,16 +5068,16 @@ var activeWorkspaceContextProvider = {
|
|
|
4795
5068
|
import * as os3 from "node:os";
|
|
4796
5069
|
import * as path5 from "node:path";
|
|
4797
5070
|
import {
|
|
4798
|
-
CredentialService,
|
|
4799
|
-
GitHubPatClient as GitHubPatClient2,
|
|
4800
|
-
MemoryTokenStore,
|
|
4801
|
-
WorkspaceService
|
|
5071
|
+
CredentialService,
|
|
5072
|
+
GitHubPatClient as GitHubPatClient2,
|
|
5073
|
+
MemoryTokenStore,
|
|
5074
|
+
WorkspaceService
|
|
4802
5075
|
} from "git-workspace-service";
|
|
4803
5076
|
|
|
4804
5077
|
// src/services/workspace-github.ts
|
|
4805
5078
|
import {
|
|
4806
|
-
GitHubPatClient,
|
|
4807
|
-
OAuthDeviceFlow
|
|
5079
|
+
GitHubPatClient,
|
|
5080
|
+
OAuthDeviceFlow
|
|
4808
5081
|
} from "git-workspace-service";
|
|
4809
5082
|
function parseOwnerRepo(repo) {
|
|
4810
5083
|
const match = repo.match(/(?:github\.com\/)?([^/]+)\/([^/.]+)/);
|
|
@@ -4860,7 +5133,9 @@ async function performOAuthFlow(ctx, clientId) {
|
|
|
4860
5133
|
expiresIn: deviceCode.expiresIn
|
|
4861
5134
|
});
|
|
4862
5135
|
} else {
|
|
4863
|
-
console.log(
|
|
5136
|
+
console.log(`
|
|
5137
|
+
[GitHub Auth] Go to ${deviceCode.verificationUri} and enter code: ${deviceCode.userCode}
|
|
5138
|
+
`);
|
|
4864
5139
|
}
|
|
4865
5140
|
const token = await oauth.pollForToken(deviceCode);
|
|
4866
5141
|
const client = new GitHubPatClient({ token: token.accessToken });
|
|
@@ -4929,7 +5204,8 @@ async function getStatus(workspacePath) {
|
|
|
4929
5204
|
cwd: workspacePath,
|
|
4930
5205
|
encoding: "utf-8"
|
|
4931
5206
|
}).trim();
|
|
4932
|
-
const lines = statusOutput.split(
|
|
5207
|
+
const lines = statusOutput.split(`
|
|
5208
|
+
`).filter(Boolean);
|
|
4933
5209
|
const modified = [];
|
|
4934
5210
|
const staged = [];
|
|
4935
5211
|
const untracked = [];
|
|
@@ -5637,6 +5913,7 @@ async function handleAgentRoutes(req, res, pathname, ctx) {
|
|
|
5637
5913
|
}
|
|
5638
5914
|
|
|
5639
5915
|
// src/api/coordinator-routes.ts
|
|
5916
|
+
var COORDINATOR_PREFIX = "/api/coding-agents/coordinator";
|
|
5640
5917
|
async function handleCoordinatorRoutes(req, res, pathname, ctx) {
|
|
5641
5918
|
if (!pathname.startsWith(COORDINATOR_PREFIX)) {
|
|
5642
5919
|
return false;
|
|
@@ -5654,7 +5931,9 @@ async function handleCoordinatorRoutes(req, res, pathname, ctx) {
|
|
|
5654
5931
|
"Cache-Control": "no-cache",
|
|
5655
5932
|
Connection: "keep-alive"
|
|
5656
5933
|
});
|
|
5657
|
-
res.write(
|
|
5934
|
+
res.write(`:ok
|
|
5935
|
+
|
|
5936
|
+
`);
|
|
5658
5937
|
const unsubscribe = coordinator.addSseClient(res);
|
|
5659
5938
|
req.on("close", unsubscribe);
|
|
5660
5939
|
const keepAlive = setInterval(() => {
|
|
@@ -5662,7 +5941,9 @@ async function handleCoordinatorRoutes(req, res, pathname, ctx) {
|
|
|
5662
5941
|
clearInterval(keepAlive);
|
|
5663
5942
|
return;
|
|
5664
5943
|
}
|
|
5665
|
-
res.write(
|
|
5944
|
+
res.write(`:ping
|
|
5945
|
+
|
|
5946
|
+
`);
|
|
5666
5947
|
}, 30000);
|
|
5667
5948
|
req.on("close", () => clearInterval(keepAlive));
|
|
5668
5949
|
return true;
|
|
@@ -5747,7 +6028,6 @@ async function handleCoordinatorRoutes(req, res, pathname, ctx) {
|
|
|
5747
6028
|
}
|
|
5748
6029
|
return false;
|
|
5749
6030
|
}
|
|
5750
|
-
var COORDINATOR_PREFIX = "/api/coding-agents/coordinator";
|
|
5751
6031
|
|
|
5752
6032
|
// src/api/issue-routes.ts
|
|
5753
6033
|
async function handleIssueRoutes(req, res, pathname, ctx) {
|
|
@@ -5979,6 +6259,7 @@ async function handleWorkspaceRoutes(req, res, pathname, ctx) {
|
|
|
5979
6259
|
}
|
|
5980
6260
|
|
|
5981
6261
|
// src/api/routes.ts
|
|
6262
|
+
var MAX_BODY_SIZE = 1024 * 1024;
|
|
5982
6263
|
async function parseBody(req) {
|
|
5983
6264
|
return new Promise((resolve5, reject) => {
|
|
5984
6265
|
let body = "";
|
|
@@ -6035,7 +6316,6 @@ function createCodingAgentRouteHandler(runtime, coordinator) {
|
|
|
6035
6316
|
};
|
|
6036
6317
|
return (req, res, pathname) => handleCodingAgentRoutes(req, res, pathname, ctx);
|
|
6037
6318
|
}
|
|
6038
|
-
var MAX_BODY_SIZE = 1024 * 1024;
|
|
6039
6319
|
|
|
6040
6320
|
// src/index.ts
|
|
6041
6321
|
var codingAgentPlugin = {
|
|
@@ -6080,5 +6360,5 @@ export {
|
|
|
6080
6360
|
CodingWorkspaceService
|
|
6081
6361
|
};
|
|
6082
6362
|
|
|
6083
|
-
//# debugId=
|
|
6363
|
+
//# debugId=7296EE1D6046456D64756E2164756E21
|
|
6084
6364
|
//# sourceMappingURL=index.js.map
|