marketing-shibata50/multica
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

refactor(subagent): improve system prompt with anti-hallucination and group guidance

Browse source 
Rewrite Sub-Agents section: add explicit anti-hallucination rules (never
fabricate completion status), make group/next the recommended pattern for
multi-agent tasks, remove sessions_list polling restrictions, add usage
examples.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
Jiang Bohan 2026-02-12 17:11:38 +08:00
parent 48185c3e77
commit 6b9e5d5f40
1 changed files with 35 additions and 11 deletions
Download patch file Download diff file Expand all files Collapse all files

46
packages/core/src/agent/system-prompt/sections.ts
View file

@ -262,23 +262,47 @@ export function buildConditionalToolSections(
lines.push(
"## Sub-Agents",
"If a task is complex or long-running, spawn a sub-agent. It will do the work and report back when done.",
"IMPORTANT: After spawning sub-agents, do NOT immediately check on them with sessions_list. " +
"Results are delivered directly into your context automatically when the sub-agent finishes. " +
"Continue with other tasks or finish your turn and wait for the results to arrive.",
"You may use sessions_list to check on sub-agents only if a long time has passed or the user explicitly asks about their status.",
"Sub-agents cannot spawn nested sub-agents.",
"",
"### Critical Rules",
"- **NEVER fabricate, guess, or make up data that a sub-agent has not yet returned.** " +
"This includes completion status — do NOT claim tasks are done until you receive actual results.",
"- After spawning, do NOT proceed with work that depends on the sub-agent results. " +
"You can still chat with the user, do unrelated tasks, or explain what the sub-agents are working on.",
"- Sub-agents cannot spawn nested sub-agents.",
"- You can use `sessions_list` to check sub-agent status if needed.",
"",
"### Groups and Continuation (`next`) — ALWAYS use for multi-agent tasks",
"When spawning multiple sub-agents, **always** use `next` to define the follow-up work. " +
"This is the standard pattern — do NOT use bare `announce: \"silent\"` for multi-agent collect-then-act workflows.",
"",
"```",
"// First spawn — creates a group automatically, returns groupId",
'sessions_spawn({ task: "Get AAPL financials", next: "Summarize all data and write a PDF report", label: "AAPL" })',
"// → { groupId: \"grp-abc\", runId: \"...\" }",
"",
"// Subsequent spawns — join the same group",
'sessions_spawn({ task: "Get MSFT financials", groupId: "grp-abc", label: "MSFT" })',
'sessions_spawn({ task: "Get GOOG financials", groupId: "grp-abc", label: "GOOG" })',
"```",
"",
"The system waits for ALL runs in the group to complete, then delivers the combined findings " +
"plus the `next` continuation prompt back to you. You can then use tools (write files, call APIs, etc.) " +
"to complete the follow-up work. The user is NOT blocked during this process — they can keep chatting.",
"",
"Use `next` whenever the user's request involves: collect data → then act on it (summarize, analyze, generate files).",
"Without `next`, findings are summarized but no further action is taken.",
"",
"### Announce Modes (when not using groups)",
"- `announce: \"immediate\"` (default): findings delivered per sub-agent as each completes.",
"- `announce: \"silent\"`: all findings held until every silent sub-agent finishes, then delivered together.",
"Groups always use silent collection internally — you don't need to set announce when using groupId.",
"",
"### Timeout Guidelines",
"Set timeoutSeconds generously — a sub-agent that times out loses all its work.",
"- Simple tasks (search, read, summarize): 600 (10 min, the default)",
"- Moderate tasks (multi-step research, file downloads + analysis): 900–1200 (15–20 min)",
"- Complex tasks (code generation, PDF creation, multi-file operations): 1200–1800 (20–30 min)",
"When in doubt, use a longer timeout. It is always better to wait longer than to lose completed work.",
"",
"### Announce Modes",
"- `announce: \"immediate\"` (default): Each sub-agent's findings are delivered to you as soon as it completes.",
"- `announce: \"silent\"`: All findings are held back until every silent sub-agent finishes, then delivered as ONE combined report.",
"Use \"silent\" when you want to collect data from multiple sub-agents first, then summarize everything at once.",
"When in doubt, use a longer timeout.",
"",
);
}