start-work
Skill8 starsChanged 4 months agoMITView on GitHub
Begin or resume product work on a very-ai-product-loops instance (not onboarding — use product-setup for first run). Self-bootstraps the rules (does NOT trust CLAUDE.md auto-load, which doesn't fire when invoked as a skill or from another session), orients on the instance state, then runs the operating loop ONE pass at a time — one section produced through its method, never a bulk fill. Use at the start of any working session, on resume after a restart/compaction, or when picking up an instance someone else set up.
18 other projects use a version of this.
Adopted by
Most recently taken up by
- morochena/skillstheir own version, first seen Nov 15, 2026
- serionist/claude-session-restoretheir 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
- Spaghetticoders/developer-skillstheir own version, first seen Nov 1, 2026
- ambroselittle/agent-skillstheir own version, first seen Oct 30, 2026
Other versions18 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 18
- liferay/liferay-portal★ 2.3ksame name
- zephyrwang6/myskill★ 339same name
- notum-cz/strapi-next-monorepo-starter★ 316same name
- dannys-code-corner/incan★ 74same name
- SpaceZephyr/allSkills★ 14same name
- vantageos-agency/vantage-peers★ 4same name
- UtsavBalar1231/oh-my-claudeagent★ 3same name
- wjgoarxiv/litopencode★ 3same name
- dmzoneill/skills★ 2same name
- avantmedialtd/skills★ 1same name
8 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 repository2 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- product-setupSkill
The file
We split this skill into the rules it gives an agent
4 rules. 0 are common, 3 are distinctive, and 1 is very specific to this project.
0
Common
Most of the 14 edited versions we read keep these.
3
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 26
Do not trust auto-load: When this skill is invoked (or the repo wasn't opened as a fresh session), the framework's root CLAUDE.md was not …
- Distinctive: line 34
Read the instance's current state (state, not rules — verify it against the registers/artifacts):
- Very specific: line 71
Before context compaction, a restart, an environment change, or end of session, run the handoff operations skill …
- Distinctive: line 83
Asking technical gaps as forks: Only product decisions become 2–4 options.
“Common” is measured: kept by at least 6 in 10 of the 14 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
Start work (begin / resume a session)
The reliable entry point into the working loop. Onboarding (product-setup) sets a product up once;
this skill runs every working session after that. Its whole job is to make sure the rules are
loaded and the agent works the disciplined loop instead of bulk-filling.
Golden rule holds: the agent prepares, the human decides. One section per pass, drafts are ⚙️, real product forks are 2–4 options + a recommendation and then you wait; gaps stay
— to clarify —.
Step 0 — bootstrap the rules (before anything)
Do not trust auto-load. When this skill is invoked (or the repo wasn't opened as a fresh
session), the framework's root CLAUDE.md was not auto-loaded as the boot entry. So load the rules
yourself, in order, before touching any artifact:
process/OVERVIEW.md → OPERATING-LOOP.md → CONVENTIONS.md → REGISTERS.md. These are the
authority for everything below — this skill only walks you into them, it does not restate them.
Step 1 — Orient (read state, never guess)
Read the instance's current state (state, not rules — verify it against the registers/artifacts):
HANDOFF.md(if present) — where the last session left off. It restores state, not rules or truth; treat anything older than its last change-log entry as suspect, and run its "Environment & access" checks before relying on them.config.yaml—active_status,language,directions, source slots.sources/INDEX.md— the knowledge map; open only the sources a task needs, not the whole folder.- Determine the active status and the current step. Both are read, not guessed; if unclear, ask.
Then read only what this task needs: the current step's README.md (its gate checklist + skeleton)
and the active status's per_step[N] (goals + tool emphasis). Do not pre-load the whole framework.
Step 2 — Run ONE pass of the operating loop
OPERATING-LOOP.md is the authority; one pass produces or updates one section / gate item:
- Focus — propose the single next section/gate item (the human may redirect).
- Recommend the tool — from the status
per_steptools first, else the step default. - Check prerequisites — open that tool's
SKILL.mdundertool-skills/library/and read its prerequisites. Filling fromtemplate.mdwithout opening the method is the failure this skill exists to prevent. - Fill gaps — for each missing input, ask the human or offer to obtain it; never guess an input.
- Clarify — only real product decisions, each as 2–4 options + a ⚙️ recommendation, then wait. Technical/implementation gaps are not asked — note them as forks in the artifact.
- Act — fill the section via the tool's
template-fragment.md, tagging every claim with a source + confidence perCONVENTIONS.md; mark your own proposals ⚙️. - Update state — tick the gate item, seed/update registers (
H-…/R-…/M-…), add a dated change-log entry, surface what's still— to clarify —. - Loop or bubble — propose the next pass; if this pass invalidated a higher/lower artifact, raise it as a trigger per the step's cadence/invalidation rules.
Hard rule: one section per pass. Propose the next section and let the human steer — do not barrel through the artifact set in one go. A bulk fill bypasses the method, the prerequisites, and the human's decisions all at once (see anti-patterns).
Step 3 — At a session boundary
Before context compaction, a restart, an environment change, or end of session, run the handoff
operations skill (tool-skills/operations/handoff/) to write/update HANDOFF.md before the
boundary — so the next start-work can resume cleanly.
Anti-patterns
- Trusting auto-load. Starting work without Step 0 because "CLAUDE.md is in the repo" — it isn't in context unless it auto-loaded, which a skill invocation / cross-session start does not do.
- Bulk-filling. Producing several sections (or a whole "first cycle") in one pass. One section, one method, one pass — then propose the next.
- Template without method. Filling
template.mdsections without opening each one'sSKILL.md. - Working from the handoff alone. It restores state, not rules — Step 0 still runs first.
- Asking technical gaps as forks. Only product decisions become 2–4 options; implementation gaps are noted as forks in the artifact, not put to the human.
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.

