liuyishou

name: liuyishou description: Help workers capture work proof, protect tacit judgment, and grow a personal career moat from day-to-day work. Use when the user wants to log work, review weekly progress, spot AI-compressible tasks, generate manager/performance/internal-mobility packs, or build a local-first record of their professional value. / 帮助员工基于日常工作沉淀个人护城河。适用于记录工作、生成周度复盘、识别容易被 AI 压缩的任务、导出绩效/转岗/撤离材料。 allowed-tools: Read, Write, Edit, Bash metadata: version: "0.1.0" argument-hint: "[init|capture|weekly|moat|pack]" user-invocable: true

Language / 语言: Follow the user's language. If the user writes in Chinese, respond in Chinese. If the user writes in English, respond in English.

留一手

留一手 is a daily career-defense skill. It helps the user do real work, keep a private record of judgment and impact, and turn recent work into reusable career assets.

This skill is not for fabricating impact or sabotaging teammates. It is for:

capturing real work proof
preserving tacit knowledge before it disappears
separating shareable output from private reasoning
growing toward harder-to-replace work

When To Use

Use this skill when the user wants to:

initialize a personal career-defense workspace
capture a recent piece of work in 30-60 seconds
produce a weekly defense review
assess whether recent work is growing or shrinking their moat
export manager updates, performance bullets, mobility packs, or exit packs

Storage Model

By default, the skill stores user data under .liuyishou/ in the current workspace.

Use --root <path> only when the user explicitly wants a custom location.

Generated files:

profile.json
entries/*.json
outputs/weekly/*.md
outputs/moat/*.md
outputs/packs/*.md

Read product-shape.md when you need the product rationale. Read data-contract.md when you need the exact file layout and field meanings.

Command Map

All deterministic writes and exports should go through:

python3 ${CLAUDE_SKILL_DIR}/scripts/liuyishou.py <subcommand> ...

If CLAUDE_SKILL_DIR is unavailable, resolve the script path relative to this SKILL.md.

1. `init`

Use when .liuyishou/profile.json does not exist yet, or the user wants a fresh setup.

Required fields:

current role
current work focus
current goal

Optional fields:

company or team alias
level or years
target roles
soft skills to grow
privacy boundary
extra notes

Ask only for missing required fields. Do not force a long onboarding form.

Command pattern:

python3 ${CLAUDE_SKILL_DIR}/scripts/liuyishou.py init \
  --root .liuyishou \
  --role "后端工程师" \
  --focus "支付与结算" \
  --goal "保住当前位置" \
  --company "某大厂" \
  --level "P6" \
  --target-role "技术负责人" \
  --soft-skill "跨团队推进" \
  --privacy-boundary "分享结果与流程，保留判断脉络和敏感上下文"

After running init, summarize the created profile in one short paragraph.

2. `capture`

Use when the user mentions a recent task, incident, launch, review, alignment, rescue, or decision.

Required inputs:

what happened
why it mattered

Optional inputs:

impact
evidence link
stakeholders
workstream
tags
private notes
share level

Default behavior:

if the user already gave enough detail, do not interrogate them
if details are sparse, ask only for the two required fields
keep capture fast

Signal assessment

Before calling the capture script, assess the entry using all available context (summary, importance, impact, private notes, the user's profile). Do not rely on keyword matching — use your judgment.

Moat dimensions — pick all that genuinely apply (comma-separated):

judgment: the work required making tradeoffs, scoping decisions, prioritization, or risk assessment — not just executing a spec
influence: the work required persuading people, aligning stakeholders, driving cross-team action, or resolving conflict
domain-context: the work depended on knowledge only someone with lived experience would have — incident history, legacy quirks, edge cases, undocumented behavior
leverage: the work produced something reusable — a tool, automation, template, or process — that amplifies future output
mobility: the work demonstrates signals toward a next-level or adjacent role — mentoring, presenting, planning roadmaps, hiring
execution-only: the work was straightforward execution with no meaningful judgment, influence, or domain depth — use this when none of the above apply

AI compression risk — pick one:

high: an AI tool could do 80%+ of this work today (data reformatting, copy-pasting between systems, simple summarization, template-filling)
medium: AI can assist but a human's judgment, taste, or context is still the bottleneck
low: the work is deeply dependent on human judgment, interpersonal dynamics, organizational context, or domain intuition — AI cannot meaningfully replace it in the near term

Share advice — write one sentence of concrete guidance:

Consider the share_level, whether private_notes contain sensitive content, and the user's privacy_boundary from their profile
If content involves people-sensitive context (performance, politics, reorgs, interpersonal conflict), advise keeping dynamics private
If the work is high AI-risk, advise highlighting the judgment and quality bar that made the work matter
Otherwise, advise sharing the outcome and keeping the reasoning private

Pass your assessment via the --moat-dimensions, --ai-risk, and --share-advice flags.

Command pattern:

python3 ${CLAUDE_SKILL_DIR}/scripts/liuyishou.py capture \
  --root .liuyishou \
  --summary "修复了结算对账的漏单问题" \
  --importance "避免月末人工兜底和客户投诉" \
  --impact "把人工排查时间从半天降到二十分钟" \
  --evidence "https://example.com/pr/123" \
  --stakeholder "财务" \
  --stakeholder "客服" \
  --workstream "结算稳定性" \
  --tag "incident" \
  --tag "cross-team" \
  --private-note "真正难点是历史订单状态不一致，公开材料里不要展开" \
  --share-level "manager" \
  --moat-dimensions "judgment,domain-context" \
  --ai-risk "low" \
  --share-advice "Share the fix outcome and timeline; keep the root cause analysis about legacy state inconsistency in your private vault."

After capture:

confirm what was stored
surface the moat signals in plain language — explain why you assigned each dimension and risk level
suggest one thing worth adding if the entry lacks evidence or judgment

3. `weekly`

Use when the user asks for:

weekly review
周报
这周我做了什么
本周护城河复盘

Command pattern:

python3 ${CLAUDE_SKILL_DIR}/scripts/liuyishou.py weekly --root .liuyishou --days 7

Then read the generated markdown and also read the entry files in .liuyishou/entries/ for the same period.

Return:

one short summary
the top value proof from the week
the top moat risk from the week
the top 3 actions for next week

Rewriting next actions

The script generates placeholder next actions using simple rules. After reading the report and the raw entries, replace them with your own assessment. Consider:

the user's profile (role, focus, goal, target roles)
which moat dimensions are strong vs missing
the AI compression risk mix — if too many entries are high-risk, suggest pulling toward harder-to-replace work
what specific, actionable steps would improve the user's position next week
each action should be distinct and concrete, not generic filler

Use the Edit tool to replace the ## Next Week Actions section in the generated markdown file, then present the updated version to the user.

4. `moat`

Use when the user wants a broader 30-day assessment of replaceability, growth direction, or AI-compression risk.

Command pattern:

python3 ${CLAUDE_SKILL_DIR}/scripts/liuyishou.py moat --root .liuyishou --days 30

Then read the generated report and also read the entry files in .liuyishou/entries/ for the same period.

Explain:

where the user's recent value is coming from
what work looks too commoditized
what kind of work they should pull toward next

Rewriting recommended moves

The script generates placeholder recommended moves using simple rules. After reading the report and the raw entries, replace them with your own assessment. Consider:

the user's profile and career goal
which moat dimensions dominate and which are absent
whether the user is drifting toward easily-automated work
what specific shifts in work selection would strengthen their moat over the next 30 days
each move should be distinct and tied to real patterns you see in the entries

Use the Edit tool to replace the ## Recommended Moves section in the generated markdown file, then present the updated version to the user.

5. `pack`

Use when the user wants a reusable artifact for someone else.

Supported pack types:

manager-update
performance-bullets
mobility-pack
exit-pack

Command patterns:

python3 ${CLAUDE_SKILL_DIR}/scripts/liuyishou.py pack --root .liuyishou --type manager-update
python3 ${CLAUDE_SKILL_DIR}/scripts/liuyishou.py pack --root .liuyishou --type performance-bullets --days 90
python3 ${CLAUDE_SKILL_DIR}/scripts/liuyishou.py pack --root .liuyishou --type mobility-pack --days 90
python3 ${CLAUDE_SKILL_DIR}/scripts/liuyishou.py pack --root .liuyishou --type exit-pack --days 90

pack is intentionally split in two layers:

the script writes a deterministic local draft
the agent rewrites the audience-facing judgment and wording

The script must exclude share_level = private entries from all pack exports by default. Never reintroduce private entries or private_notes in your rewrite.

After generation, read the markdown file and the raw shareable entries.

For manager-update, rewrite the draft into a short manager-facing weekly update. Default shape:
- ## Wins: 3-5 flat bullets, ordered by strategic importance
- ## Risks / Asks: 0-2 flat bullets
- ## Next Steps: exactly 3 items
- Do not add long prose, capability essays, or theme subheadings unless the user explicitly asks.
For performance-bullets, default to 5-8 copy-ready Chinese bullets only. Each bullet should follow "做了什么 + 为什么重要 + 结果/影响". Do not add sections, tables, maturity models, keyword indexes, or long commentary unless the user explicitly asks for a longer review packet.
For mobility-pack, rewrite around target-role fit. A structured memo is fine here: role target, ability mapping, strongest evidence, gaps, next proof points.
For exit-pack, you may keep the longer emergency-switch format: short self-intro, transferable stories, resume-ready bullets, target-role directions, evidence index.

Operating Rules

Never invent achievements, numbers, or evidence.
Distinguish clearly between:
- shareable facts
- private reasoning
- people-sensitive context
Prefer deterministic script writes for storage, filtering, and scaffolding; use LLM judgment for audience-specific wording and recommendations.
Keep user data local unless the user explicitly asks for export or sharing.
If the user asks for a new feature, preserve the existing storage contract in data-contract.md.

Minimal Intake Rules

For fast adoption, keep prompts short.

init needs only:

role
focus
goal

capture needs only:

what happened
why it mattered

Everything else is optional and can be enriched later.

Output Contract

This skill should produce two classes of output:

Private outputs:

work receipts timeline
weekly defense review
moat report

Shareable outputs:

manager update
performance bullets
mobility pack
exit pack

Output formats for v0:

markdown files on disk
short in-chat summary
copy-ready bullets

Artifact shape defaults:

manager-update: short enough to paste into chat or email without trimming
performance-bullets: bullets first, no report styling by default
mobility-pack: argument memo for internal role change
exit-pack: external-facing interview and resume ammo pack

Do not add PDF, database, browser extension, or cloud sync logic inside this skill unless the user explicitly asks for it.

留一手

When To Use

Storage Model

Command Map

1. init

2. capture

Signal assessment

3. weekly

Rewriting next actions

4. moat

Rewriting recommended moves

5. pack

Operating Rules

Minimal Intake Rules

Output Contract

1. `init`

2. `capture`

3. `weekly`

4. `moat`

5. `pack`