start-work
Skill1 starsChanged 16 days agoMITView on GitHub
Starts a piece of work from a list of tickets (Jira, Linear, ClickUp) or from a description. Reads the cards and their comments, writes a plan and stops to wait for the go-ahead. Then runs the full protocol (branch, development, tests, gui-test, audit rounds, report) without committing until told to.
17 other projects use a version of this.
Adopted by
Most recently taken up by
- serionist/claude-session-restoretheir own version, first seen Nov 8, 2026
- timamikh/very-ai-product-loopstheir own version, first seen Nov 8, 2026
- elpiarthera/vantage-startertheir own version, first seen Nov 1, 2026
- spalladino/claude-skillstheir own version, first seen Nov 1, 2026
- ambroselittle/agent-skillstheir own version, first seen Oct 30, 2026
- dmzoneill/skillstheir own version, first seen Oct 24, 2026
Other versions17 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 17
- liferay/liferay-portal★ 2.3ksame name
- zephyrwang6/myskill★ 339same name
- notum-cz/strapi-next-monorepo-starter★ 316same name
- dannys-code-corner/incan★ 73same name
- SpaceZephyr/allSkills★ 14same name
- timamikh/very-ai-product-loops★ 8same name
- vantageos-agency/vantage-peers★ 4same name
- UtsavBalar1231/oh-my-claudeagent★ 3same name
- dmzoneill/skills★ 2same name
- wjgoarxiv/litopencode★ 2same name
7 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository5 other files it gives its agents.
- deliverSkill
- figma-checkSkill
- gui-testSkill
- squadraSkill
- verify-for-realSkill
The file
We split this skill into the rules it gives an agent
6 rules. 0 are common, 5 are distinctive, and 1 is very specific to this project.
0
Common
Most of the 13 edited versions we read keep these.
5
Distinctive
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 16
Check whether it's already done: Before estimating, look at the code: a card sometimes describes something someone else has already closed.
- Distinctive: line 26
Then stop and wait for the go-ahead: Don't touch the code before that.
- Distinctive: line 28
If reading raises a question that would change the plan, ask it now, not halfway through the work.
- Distinctive: line 39
Cases that can't be produced on demand get mocked: If a condition needs data the backend doesn't return (an empty profile, an expired …
- Very specific: line 54
Every fix must be verified with verify-for-real: break it and watch the test fail before keeping it.
- Distinctive: line 59
Stop and ask before implementing: Don't invent copy or measurements, and don't silently implement the version that looks right to you.
“Common” is measured: kept by at least 6 in 10 of the 13 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Starting a piece of work
Invoked with a list of tickets, or with a description in words. Everything else in this file is the protocol: it applies even when the user doesn't repeat it.
1. Understand before planning
- Read every card and its comments. Comments hold course corrections the description doesn't: fix requests, decisions made in a call, Figma nodes. A card read without its comments is a card read halfway.
- Check whether it's already done. Before estimating, look at the code: a card sometimes describes something someone else has already closed. Checking costs a grep and changes the plan.
- Project facts (base branch, tracker, simulator, flavor, Figma file) live in
the project docs:
AGENTS.md,CLAUDE.md, the project memory. Read them, don't ask for them. Only ask for what really isn't there.
2. The plan, and then stop
A short plan: what changes, where, what breaks, what is still to be decided. Then stop and wait for the go-ahead. Don't touch the code before that.
If reading raises a question that would change the plan, ask it now, not halfway through the work.
3. Execution
Branch from the project's base branch, following the convention the repo
already uses (look at the latest branches: git for-each-ref --sort=-committerdate refs/remotes).
Then, for each slice: development → tests → gui-test on the simulator →
iterate until the screen is right. See gui-test and figma-check.
Cases that can't be produced on demand get mocked. If a condition needs data the backend doesn't return (an empty profile, an expired cache, no network), build a mock page in a developer menu, next to the project's other ones. It's also how people who don't code get to try the case.
4. Audit rounds
Audit 1, 2, ... N. An audit means rereading the diff and rerunning tests and gui-test, not just looking at the code.
Stop when a whole round produces only cosmetic findings: formatting, a stale number in a doc. That's the sign the curve has flattened: the first rounds find real defects, the last ones hygiene. Then one last security audit (what ends up on disk, what in the logs, what survives logout) and one last round of tests.
Every fix must be verified with verify-for-real: break it and watch the test
fail before keeping it.
5. When the design looks wrong
Stop and ask before implementing. Don't invent copy or measurements, and don't silently implement the version that looks right to you.
But don't block everything: do all the work that doesn't depend on that answer, hold back only the uncertain piece, and ask at the right moment, not at the end of the turn. If the user decides to deviate from Figma, the deviation must be written down (code, docs, test) or someone will "restore" it by rereading the node.
6. Wrapping up
- No commits and no pushes until the user says so. See
deliver. - No comments on the tickets unless asked.
- A final report: what was done, which defects came up and how they were measured, what is still open and what is waiting for the user's decision.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

