English | 中文
An evidence-driven weekly planning Skill for work that needs real-world feedback.
Proofweek turns vague intentions into a small, editable plan with preparation limits, early external action, objective completion evidence, and an explicit next decision. It is designed for moments when research, polishing, or organizing keeps displacing applications, publishing, user tests, or outreach.
Proofweek reduces the cost of planning and makes preparation-heavy avoidance visible. It does not diagnose procrastination, guarantee behavior change, or treat activity as proof of demand.
A normal task list can make both of these lines look equally productive:
- refine the résumé again;
- apply to a real role and record what happened.
But they answer different questions. The first may improve readiness; only the second creates contact with the outside world. Proofweek keeps those stages separate, limits preparation, and brings the first real test into the current planning period.
Typical situations include:
- Job search: résumé and portfolio polishing keeps delaying real applications.
- Content publishing: research, wording, and cover design keep delaying publication.
- MVP testing: feature work and experience improvements keep delaying contact with users.
- Outreach or acquisition: list building and message refinement keep delaying actual outreach.
Proofweek is not intended for ordinary errands, household checklists, purely internal engineering schedules, long-term life planning, or work that does not need an external judgment.
Proofweek separates a weekly plan into five kinds of work:
| Type | Question it answers | Acceptable evidence | What it does not prove |
|---|---|---|---|
| Preparation | What must exist before submission? | A file, candidate list, material version, or bounded research conclusion | That anyone will accept it |
| Technical QC | Is the minimum path usable and factually correct? | A reproducible check, test record, or usable version | Market demand or recruiter interest |
| Market sampling | Is a small batch worth expanding? | A completed batch of applications, posts, invitations, or outreach records | Stable demand or broad acceptance |
| External validation | What did a real person actually say or do? | A reply, quote, rejection, observed behavior, or conversion | What unobserved people would do |
| Decision | What changes because of the evidence? | A written continue, modify, stop, or wait decision with its basis | That the decision is certainly correct |
The Skill then:
- extracts facts, constraints, available time, materials, deadlines, and external audience from the current conversation;
- proposes a task hypothesis before asking questions;
- asks only one decisive question when an answer would change the core result, first external action, evidence, or a hard blocker;
- rewrites each row as one action plus one checkable completion record;
- caps preparation by time, quantity, or deadline;
- schedules a small external action early instead of placing every test after all preparation;
- produces an editable Markdown weekly plan, weekend review, and continuation block.
There is deliberately no “reason I am unwilling to submit” field. Proofweek checks factual blockers rather than debating motivation.
Vague input
Update my résumé, improve my projects, select AI product manager roles, and start applying this week.
Proofweek changes the plan into a sequence like this:
- Preparation: select a bounded set of matching roles; evidence = saved role list and requirements matrix.
- Technical QC: verify facts, links, formatting, and required fields in the résumé; evidence = named résumé version plus QC record.
- Market sampling: submit a small first batch before further portfolio expansion; evidence = application records.
- External validation: record screening, interview, rejection, or no-response outcomes without interpreting them prematurely.
- Decision: continue the résumé version, revise one hypothesis, or change role selection.
An application record proves that an application happened. It does not prove that the résumé is accepted or that the role is a match.
Instead of combining “research the topic, learn posting tactics, improve the cover, and publish” into one row, Proofweek sets a preparation budget, checks the minimum publishable version, and schedules the post while the topic is still timely. The result field records the link and observed responses. Likes are evidence of likes—not automatic proof of purchase intent or durable demand.
Proofweek first separates real blockers from optional polish. A privacy, confidentiality, data-deletion, factual, or broken-core-flow issue can stop a test. Extra features and general “it could be better” work cannot silently postpone it. After the minimum path is usable, the plan invites a small, named group and records what those users actually do or say.
If the Skill cannot identify a real role, platform, customer, tester, or other outside evaluator from the conversation, it asks exactly one question and stops:
这次要先让哪类真实对象评价或使用什么？
It does not fill the plan with placeholders such as “target users” and pretend the external test is defined.
Install from GitHub with the Agent Skills installer:
npx -y skills add RuntianLee/proofweek -g --allRequirement: a client that supports Agent Skills.
After installation, invoke the Skill by name:
Use $proofweek to turn my weekly intentions into an editable plan that limits preparation and obtains external evidence early.
Useful input, when available:
- what you want to move forward this week;
- available time;
- existing materials or current progress;
- the real people, roles, platforms, customers, or testers you can reach;
- a deadline;
- known factual blockers, or
none.
You do not need to complete a form before starting. Proofweek uses information already present in the conversation and only asks when a missing answer would materially change the plan.
A weekly plan contains two layers: a short definition of success, followed by an execution table.
### Core result
- Deadline: Friday 18:00
- Controllable action: submit the QC-passed résumé to a first batch of matching roles
- External audience: selected AI product manager openings
- Completion evidence: résumé version + application records
### Constraints
- Available capacity: 6 hours
- Preparation budget: 2 hours or one résumé version
- First external action deadline: Wednesday 20:00
- Known hard blockers: none| Status | Action + completion evidence | Goal | Type | Planned period | Actual period | Objective result → next step |
|---|---|---|---|---|---|---|
| To do | QC résumé v3; evidence = QC record + saved file | Result 1 | Technical QC | Tue 20:00–21:00 | — | — |
| To do | Submit first batch; evidence = application records | Result 1 | Market sampling | Wed 20:00–21:00 | — | — |
Planned and actual periods may contain multiple work blocks when they all serve one final piece of evidence:
09-03 20:00–21:00
09-04 21:30–22:15
Σ 105m
Split the row when an intermediate step has its own independently acceptable deliverable.
- Weekly planning: turn intentions, current materials, and constraints into a plan.
- Weekend review: separate completed actions, technical QC, external evidence, unknowns, and decisions.
- Cross-session continuation: paste the generated continuation block into a new conversation to restore the current facts and commitments.
Proofweek prefers records that another person can inspect:
- a file or recoverable version;
- a link, screenshot, send, application, publication, or invitation record;
- a reproducible test result;
- a real person's words or observable behavior;
- a written decision with its evidence.
The following distinctions are intentional:
- an application record ≠ résumé acceptance;
- a technical pass ≠ user need;
- a like ≠ purchase intent;
- free usage ≠ willingness to pay;
- one person's outcome ≠ a shared market need;
- no response = an observed outcome, not automatic rejection or acceptance.
- Proofweek never applies, publishes, contacts people, or modifies third-party systems without explicit authorization.
- Privacy, compliance, confidentiality, factual errors, irreversible risk, or a broken core flow can legitimately block external action.
- The Skill does not invent dates, time estimates, links, feedback, or audiences.
- It does not provide psychological diagnosis or medical claims.
- When no external judgment is needed, it says the workflow is not applicable instead of manufacturing “market validation.”
proofweek/
├── README.md
├── README.zh-CN.md
├── LICENSE
├── assets/
│ └── proofweek-social-preview.png
└── skills/
└── proofweek/
├── SKILL.md
├── agents/openai.yaml
├── assets/weekly-plan-template.md
└── references/decision-rules.md
Only the files required to run the Skill are published in the installable package. Local evaluation samples, expected answers, and run records are intentionally excluded.
Completed so far:
- Skill project structure validation;
- five independent behavior samples covering job search, content publishing, a missing external audience, an out-of-scope task, and a privacy blocker;
- installation verification through
npxin an isolated directory; - use of the workflow by the author in a real weekly-planning process.
Still unknown:
- whether other people repeatedly experience and describe the same planning problem;
- whether the workflow improves external-action timing across different domains;
- whether any effect persists over multiple weeks;
- whether users prefer a Skill, a document template, or integration with an existing task tool.
Technical validity and one person's usage are not treated as product-market evidence.
Issues and pull requests are welcome. Useful contributions include:
- a reproducible case where the Skill asks unnecessary questions;
- a plan that confuses preparation with external evidence;
- an example where a legitimate blocker is ignored;
- compatibility notes for Agent Skills clients;
- clearer wording or bilingual documentation fixes.
Please remove private job, customer, user, or company information from examples before sharing them.