SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsskill-plan

skill-plan

Writing & research

View repository

Updated 10-16 08:39

About this skill

/skill-plan — Personalized learning strategy (gap → path → landing a job)

Turn a target role into an executable learning plan that is "based on your starting point + weekly time, learnable to the point of being able to submit resumes within a few months."

Pre-checks

  • Read .skill-state.json to find the role. If it is not in candidate_roles or its learnability assessment is ❌ discouraged/⛔ too difficult,first remind the user to go back to skill-scan to reselect or confirm, and do not force a plan that is doomed to fail.
  • If the role involves paid training/referrals, first confirm that it passes ../../shared-references/anti-scam-rubric.md; if it hits a red line, discourage it first.
  • Read ../../shared-references/learnability-rubric.md(gap dimensions aligned) and role-tiers.md(starting actions by tier).

The learning strategy must include 6 blocks

  1. Gap analysis (most important): target role JD requirements − the user's existing abilities = the list to fill, listed item by item,mark which are transferable (already have them, only need slight learning) and which must be learned from scratch. The gap list determines the entire plan; do not skip it.

  2. Phased learning path: break it into 2–4 phases, each phase having:

    • Goal (what you can do after learning it, verifiable)
    • Specific content + recommended resource types (prioritize free/low-cost + learn by doing; when recommending specific courses/resources, use WebSearch to check what is actually available now, add the year, and do not give links from memory that may be invalid)
    • AI acceleration points: how to use AI in this phase as a private tutor / use AI programming / use AI to produce work, to shorten the learning curve (this is the key point of learning in the AI era; do not let "using AI" remain a slogan)
    • Time box (calculated based on the user's N hours per week; how many weeks this phase takes)

  1. Portfolio checklist: list 2–4pieces of work that can go on a resume/be demonstrated(by tier: S0 tool proficiency work; S1 old skills × AI redone; S2 runnable GitHub projects; S3 demos that solve real pain points in this industry). Work is worth more than certificates—it proves you can really do it.

  2. Job search timeline: which week to start submitting resumes, which channels to use, and what to prepare for interviews. Make clear that "you do not have to wait until you have learned everything before applying"—apply while filling gaps.

  3. Stop-loss line: what conditions mean you should stop, review, or adjust (e.g., 30 applications with 0 interviews / after X weeks you are completely unable to keep up / anti-fraud red lines appear). Leave a clear-eyed exit judgment for a career transition.

  4. Honest timeline and expectations: based on the user's weekly time, how many months until they can apply, and how many months until they may get an offer. State clearly that this is real investment, not a quick fix; do not paint a rosy picture.

When needed, use WebSearch to check the role's current real learning path/mainstream tool stack/hiring requirements (add the year); do not give outdated plans.

Save to disk (including "expectations"—this is the starting point for later review and reconciliation)

Before writing, first get the current time using the system time:

date '+%Y-%m-%d %H:%M %Z %z'

All reports and status records must include a specific time and time zone, following the user's system time zone. Internal records may use 2026-06-26 22:06 CST +0800; user reports should be displayed as 2026-06-26 22:06（本机时区 + UTC 偏移，如 中国标准时间 CST，UTC+8）. Do not hardcode it to the time of any fixed country.

Write the strategy into .skill-state.json of active(chosen_id / started_at=today / learning_plan), and change that role's status to learning。 Also write active.prediction: estimated weekly hours of commitment, estimated number of months until you can apply with a resume, portfolio checklist, stop-loss line (recorded_at=today's date, recorded_at_full=specific time and time zone). Once this expectation is written, do not change it—when you review later, compare the "actual" against it to see clearly whether the plan was accurate and whether it needs adjustment.

Report and archive wrap-up

After the learning plan is generated, do not end directly. First ask the user:

"If this version of the direction and plan is fine, I can help you archive it and generate a complete report. The recommended format is a Markdown source file + an optional Word version: Markdown is convenient for later iteration, and Word is convenient for sending/printing. Do you want to generate it now?"

By default, do not use Markdown/HTML as user deliverables. For ordinary users, seeing code or markup syntax is confusing.

If the user agrees, generate two types of content:

  • Internal archive: continue writing into .skill-state.json; if necessary, generate reports/internal/<YYYY-MM-DD>-<target-slug>.md as a structured archive that the agent can maintain, but do not use it as the main deliverable to the user.
  • User-readable source draft: first generate reports/<YYYY-MM-DD>-<中文标题>-用户版.txt, and write a complete report in natural language that even a layperson can understand. Do not include HTML/CSS/Markdown syntax, and do not ask the user to open it in a browser. The report body must include 生成时间：YYYY-MM-DD HH:mm（本机时区名 缩写，UTC 偏移）, following the user's system time zone. Do not write only the date, and do not hardcode the time of any country. Example:生成时间：2026-06-26 22:06（中国标准时间 CST，UTC+8）。

Export and delivery rules:

  • By default, prioritize exporting the user-readable version as reports/<YYYY-MM-DD>-<中文标题>-用户版.docx。
  • If .docx export succeeds, only tell the user the Word file save path, and open the folder containing the document so the user can double-click the Word file themselves; do not additionally generate or display .rtf / .txt as the main deliverable. You may add: "If you need plain text or RTF, I can export that too."
  • If .docx export fails, then export .txt and .rtf, tell the user the two file paths, and likewise open the containing folder.
  • Do not rely on the chat interface's file cards or Google Docs opening method; for Chinese users, deliver by default as "local folder + Word document."
  • The wrap-up wording to the user must include the save directory, for example:报告已保存到 <abs-path>/reports/，我已打开这个文件夹，你可以直接双击 Word 文档查看。
  • .docx After successful export, the final user wording should include:
    • 报告已经生成好了。
    • 保存目录：<abs-path>/reports/
    • Word 报告文件：<abs-path>/reports/<报告名>.docx
    • 我已经帮你打开了报告所在文件夹，你可以直接双击这个 Word 文档查看。
    • 这份报告里面包括： After that, list the recommended direction, recruitment information conclusion, ranking of recommended roles, roles not recommended for main focus for now, skills that need to be supplemented, learning plan, portfolio suggestions, anti-fraud and stop-loss line, and next-step actions.
    • 如果你想要 txt 版或 RTF 版，我也可以再导出。
    • 💡 后面你执行计划时，如果进度比预期快、慢，或者投递反馈和我们预判的不一样，我会根据实际情况帮你调整计划，不会让你死磕一版固定路线。

After the report is delivered, do not stop at "the file has been generated." You must naturally enter companion mode and say to the user:

"From now on, you can learn along with me. I will guide you step by step according to this plan, and you do not need to judge on your own what to learn next. Let's start with Week 1, Day 1: choose the main tool and get the first simplest AI Q&A/data organization mini-tool working."

Then immediately start the first training task, unless the user explicitly says to pause:

  1. Explain that today only 2-3 things need to be done, to reduce anxiety.
  2. Have the user choose one primary tool from 扣子/Coze and Dify , and if the user is unsure, by default recommend choosing 扣子/Cozefirst, because beginners can more easily get agents and workflows running.
  3. Give the completion standard for the day: be able to open the tool, create a simple bot/app, enter a piece of text, and get stable formatted output.
  4. Tell the user to come back after completing it and send a screenshot/description/error, and the agent will continue guiding the next step.

Suggested report structure:

  1. One-page summary: recommended direction, timeline, target salary range, primary portfolio piece.
  2. User profile: starting level, transferable skills, constraints.
  3. Hiring signals: BOSS search keywords, pages viewed, JDs read, salary level, job mode.
  4. Recommended job clusters: why choose this direction, which roles serve as anchors.
  5. Gap analysis: existing skills, skills to build, skills not to force yet.
  6. 10-week learning and application plan: phases, milestones, weekly tasks.
  7. Portfolio specifications: input, process, output, and demo materials for each piece.
  8. Job search actions: keywords, application pace, interview preparation.
  9. Stop-loss line and anti-fraud red lines.
  10. Questions for the next review.

Normal closing

  • Remind the user: execute by phase, and come back to adjust at milestones or when stuck (skill-retro is yet to be built; for now, manually update retro_log).
  • Re-emphasize the stop-loss line and anti-fraud red lines: for anything like "pay to get a job guaranteed/installment loan for learning," stop immediately.