fix: align generated skill guidance

[RerankerGuo]

Description

Fixes #1423.

Aligns generated SKILL.md guidance with the requested skill-creator practices:

• require concise two-to-four-word kebab-case names and trigger-aware descriptions
• limit frontmatter to name and description
• keep shared workflows in SKILL.md while moving variant details to references/
• avoid redundant trigger sections and standalone auxiliary documentation
• keep the OpenClaw and core generator copies aligned

No dependencies are added.

Type of change

• Bug fix (non-breaking change which fixes an issue)

How Has This Been Tested?

• Unit Test
  • cd apps/memos-local-openclaw && npm test -- --run tests/skill-prompt-policy.test.ts (3 passed)
• Test Script Or Test Steps
  • cd apps/memos-local-openclaw && npm run build
  • static policy checks against both generator copies
  • git diff --check
• Pipeline Automated Test
  • GitHub Actions passed all 16 build jobs across macOS, Ubuntu, Windows, and Python 3.10-3.13

Local limitations: npm run lint cannot start because this package does not declare or install eslint; make format cannot start because Poetry is unavailable in this environment.

Checklist

• I have performed a self-review of my own code | 我已自行检查了自己的代码
• I have commented my code in hard-to-understand areas | 我已在难以理解的地方对代码进行了注释
• I have added tests that prove my fix is effective or that my feature works | 我已添加测试以证明我的修复有效或功能正常
• I have created related documentation issue/PR in MemOS-Docs (not applicable; no documentation change)
• I have linked the issue to this PR | 我已将 issue 链接到此 PR
• I have mentioned the person who will review this PR: @bittergreen

Reviewer Checklist

• closes [Enhancement] Skill generation template does not follow skill-creator best practices #1423
• Made sure Checks passed
• Tests have been provided

[shinetata]

Approve — LGTM.

Aligns the generated SKILL.md guidance with #1423: concise 2-4 word kebab-case name, name+description-only frontmatter, progressive disclosure for multi-variant skills, and no auxiliary docs.

I looked closely at the one change that seemed risky — removing the metadata: { "openclaw": { "emoji": "{emoji}" } } line. It's safe: the {emoji} placeholder was never substituted in step1GenerateSkillMd (the replace chain only fills NAME/TITLE/SUMMARY/CONVERSATION), and no runtime code consumes that field. So this drops a dead placeholder rather than a working feature, and it satisfies the "name + description only" rule. No impact on the core memory pipeline (capture/ingest/embed/recall) — skill retrieval keys off name + description only.

Verified locally with the project-pinned vitest 2.1.9: tests/skill-prompt-policy.test.ts -> 3 passed (incl. the new case). CI is green across all 16 build jobs.

Nit (non-blocking): only the apps/memos-local-openclaw copy has the policy test; the identical packages/memos-core copy isn't covered. Consider adding a matching test or extracting a shared constant to prevent future drift between the two copies.

[Memtensor-AI]

🤖 Open Code Review

Target: PR #2100
Task: 160b279b7ccb7705
Base: dev-v2.0.24
Head: fix/skill-generator-guidance

✅ OpenCodeReview: No comments generated. Looks good to me.

Generated by cloud-assistant via Open Code Review.

[Memtensor-AI]

⚠️ Automated Test Results: ENV ISSUE

The test environment encountered an issue that requires manual attention.

Details: Test failures are caused by a SQLite migration error where the 'traces' table is missing a 'role' column, which is unrelated to the diff (which only modifies skill generator prompt templates). [advisory, non-gating] AI-generated tests on branch test/auto-gen-160b279b7ccb7705-20260716130904: 113/113 passed — these do NOT affect the PR verdict; review the branch manually.
Branch: fix/skill-generator-guidance