@gencode/agents 0.10.1 → 0.10.2

package/CHANGELOG.md

@@ -1,5 +1,11 @@
 # @gencode/agents
 
+## 0.10.2
+
+### Patch Changes
+
+- 93a0318: Agent system prompts now tell the model to place temporary and intermediate files under the workspace cleanup directories: `workspace/temp/daily/`, `workspace/temp/weekly/`, or `workspace/temp/monthly/`. These directories are created only when needed; AIMax does not pre-initialize them during startup.
+
 ## 0.10.1
 
 ### Patch Changes

package/dist/index.js

@@ -115,7 +115,7 @@ Preview:
 `)}function ih(e=[]){let t=Jm(e).trim();return t?`${t}\n`:``}function ah(){return[`## Scenes`,'If the user message contains a `<scenes category="..." type="..." items="...">...</scenes>` tag, treat it as an explicit scene activation request.',"In this tag, `category` means the scene category, `type` means the scene name, and `items` means the item list required by that scene.","The `category`, `type`, and `items` attributes may each contain one or more comma-separated values. You must split them on commas, trim whitespace, and use every requested value.","Before executing the task, you must enumerate every requested `<category, scene, item>` combination derived from those attributes.","For each requested item, you must call `skill_load(name=<item>, skillPath=/aimax_pvc/scenes/<category>/<scene>/<item>)` to load that item's `SKILL.md`.","If one `<scenes ...>` tag requests multiple items for the same scene, you must load them one by one. Do not stop after the first matching item.","Do not start implementation, analysis, tool execution, or user-facing conclusions until all required scene skills have been loaded with `skill_load`.","After loading those scene `SKILL.md` files through `skill_load`, you must follow their instructions as mandatory task-specific guidance unless a higher-priority instruction overrides them.","For scene skills, if the exact scene SKILL.md is not visible, reload it again with the original `skillPath`.","Scene skills are not loaded by default. Only load them when the user explicitly includes the `<scenes ...>` tag.","Outside explicit scene activation, normal shared skills from `<available_skills>` must still be loaded with `skill_load(name=<skill-name>)` rather than direct file reads.","When a scene is activated, you must load scene instructions only through `skill_load` with `skillPath` pointing at `/aimax_pvc/scenes/<category>/<scene>/<item>` or that directory's `SKILL.md`.","Do not load, match, or fall back to scene instructions from `<dataDir>/.aimax/skills`.","Do not load, match, or fall back to scene instructions from `/aimax/extensions/`.","Do not load, match, or fall back to scene instructions from any path declared in `<available_skills>`.",`Do not ignore, rewrite, or drop the scene tag semantics even if the natural-language request looks answerable without scene skills.`,``].join(`
 `)}function oh(){return[`## Memory Write`,`Use memory tools, not write_file/edit_file, for memory operations.`,`- memory_write: write durable long-term memory to MEMORY.md (facts, preferences, rules, decisions).`,`- memory_log: append short-lived daily/session notes to memory/YYYY-MM-DD.md or session logs.`,`- Choose exactly one write surface per fact by default: long-term -> memory_write; daily/session log -> memory_log.`,`- Do not write the same fact to both memory_write and memory_log unless the user explicitly requests dual-write.`,`- Before memory_update or memory_forget, first run memory_list or memory_search to get the exact id/path.`,`- If the user explicitly names a listed memory tool, prefer calling that exact tool.`,`- Pattern: write/log for new memory; list/search -> update/forget for existing memory.`,``].join(`
 `)}function sh(e){return[`## Memory & session recall`,`Choose the right recall surface:`,"- **Durable memory (Markdown / MEMORY.md / memory/ logs)**: stable facts, preferences, policies, decisions. Prefer `memory_search`, then `memory_get` when you need the full passage. Do not guess entry IDs or file paths.","- **Past chat transcripts across sessions** (what was said or done in earlier conversations): use `session_search`. Omit `query` to browse recent sessions; include `query` to search transcripts. Do not use `memory_search` alone to reconstruct a cross-session chat timeline.",'- **Ambiguous** ("last time" could mean chat or written notes): run `session_search` first for conversational evidence, then `memory_search` for durable notes; merge and attribute each source.',"If a listed memory tool call fails, report the failure; do not claim the tool is unavailable. Use `memory_list` before update/delete when needed.","Short tool chains: `memory_search` → `memory_get`; `memory_list` → `memory_update`; `memory_search` → `memory_forget`.",`Routing examples (intent → tool):`,'1) "上次聊天我们怎么修的" → `session_search` with keywords from the bug/symptom.','2) "团队对 release 的硬性规范是什么" → `memory_search` (durable policy).','3) "最近几个会话在忙什么" → `session_search` with no query (recent mode).','4) "把这条写进长期记忆" → `memory_write` (default), not `memory_log` or `session_search`.','5) "昨天日志里有没有记过 X" → `memory_search` on daily logs / `memory_list`.','6) "上次口头结论 vs 正式文档" → `session_search` then `memory_search`.',e===`off`?`Citations are disabled: do not include memory file paths or line numbers unless the user explicitly asks.`:`When useful, include memory evidence as Source: <path#line>.`,``].join(`
-`)}function ch(e){let t=`${e.dataDir}/.aimax`,n=`${e.dataDir}/workspace`,r=e.sandboxInfo?.hostWorkspaceDir?.trim()||n,i=e.sandboxInfo?.containerWorkspaceDir?.trim(),a=e.sandboxInfo?.enabled&&i?`For file tools, paths resolve against the cloud workspace mount ${r}. For exec commands, use paths under ${i} (or relative paths from there). Prefer relative paths whenever possible.`:`This cloud workspace is mounted user storage. Prefer relative paths to keep file tools and shell commands aligned.`,o=e.projectDir?.trim(),s=o?[`Current project directory: ${o}`,`Treat this project directory as the default cwd for repository/code work unless the user specifies another directory.`]:[];return[`## Workspace`,`Your cloud working directory is: ${r}`,...s,a,`User identity, memory, and agent state live under ${t}, not inside the cloud workspace.`,`Files such as AGENTS.md, SOUL.md, TOOLS.md, IDENTITY.md, USER.md, MEMORY.md, and the memory/ directory belong under ${t}.`,`Only create or update those files at their real .aimax paths. Never create cloud workspace copies like ${r}/AGENTS.md or ${r}/MEMORY.md unless the user explicitly asks for separate cloud documents.`,``].join(`
+`)}function ch(e){let t=`${e.dataDir}/.aimax`,n=`${e.dataDir}/workspace`,r=e.sandboxInfo?.hostWorkspaceDir?.trim()||n,i=e.sandboxInfo?.containerWorkspaceDir?.trim(),a=e.sandboxInfo?.enabled&&i?`For file tools, paths resolve against the cloud workspace mount ${r}. For exec commands, use paths under ${i} (or relative paths from there). Prefer relative paths whenever possible.`:`This cloud workspace is mounted user storage. Prefer relative paths to keep file tools and shell commands aligned.`,o=e.projectDir?.trim(),s=o?[`Current project directory: ${o}`,`Treat this project directory as the default cwd for repository/code work unless the user specifies another directory.`]:[];return[`## Workspace`,`Your cloud working directory is: ${r}`,...s,a,`User identity, memory, and agent state live under ${t}, not inside the cloud workspace.`,`Files such as AGENTS.md, SOUL.md, TOOLS.md, IDENTITY.md, USER.md, MEMORY.md, and the memory/ directory belong under ${t}.`,`Only create or update those files at their real .aimax paths. Never create cloud workspace copies like ${r}/AGENTS.md or ${r}/MEMORY.md unless the user explicitly asks for separate cloud documents.`,`Use the workspace temp directories for temporary or intermediate files:`,`- ${r}/temp/daily/ for short-lived files that can be cleaned daily.`,`- ${r}/temp/weekly/ for files that should remain available for about a week.`,`- ${r}/temp/monthly/ for files that should remain available for about a month.`,`Do not pre-initialize these directories; create them only when you actually need to write temporary files.`,``].join(`
 `)}function lh(e){if(!e?.enabled)return``;let t=[`## Sandbox`,`You are running in a containerized cloud CLI runtime.`];return e.containerWorkspaceDir?.trim()&&t.push(`Container workdir: ${e.containerWorkspaceDir.trim()}`),e.hostWorkspaceDir?.trim()&&t.push(`Host-mounted workspace: ${e.hostWorkspaceDir.trim()}`),t.push(``),t.join(`
 `)}function uh(e){let t=[];if(e.timezone&&t.push(`Time zone: ${e.timezone}`),e.currentDate){t.push(`Current date: ${e.currentDate}`);let n=dh(e.currentDate);n&&t.push(`Day of week: ${n}`)}return t.length===0?``:[`## Current Date & Time`,...t,``].join(`
 `)}function dh(e){let t=/^(\d{4})-(\d{2})-(\d{2})$/.exec(e.trim());if(!t)return null;let n=Number(t[1]),r=Number(t[2]),i=Number(t[3]),a=new Date(Date.UTC(n,r-1,i));return Number.isNaN(a.getTime())||a.getUTCFullYear()!==n||a.getUTCMonth()!==r-1||a.getUTCDate()!==i?null:[`Sunday`,`Monday`,`Tuesday`,`Wednesday`,`Thursday`,`Friday`,`Saturday`][a.getUTCDay()]}function fh(e){if(e.isMinimal||!e.enabled)return``;let t=(e.channels??[]).map(e=>e.trim()).filter(Boolean);return[`## Messaging`,`Reply in the current session for user-visible responses.`,`Use subagent tools only when the user or system instructions explicitly require subagents, delegation, or parallel agent execution; otherwise stay in the current agent.`,`Do not use shell/network commands to bypass the platform messaging pipeline.`,t.length>0?`Configured channels: ${t.join(` | `)}`:``,``].filter(Boolean).join(`

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@gencode/agents",
-  "version": "0.10.1",
+  "version": "0.10.2",
   "type": "module",
   "main": "./dist/index.js",
   "types": "./dist/index.d.ts",