very-ai-product-loops
CLAUDE.md8 starsChanged 4 months agoMITView on GitHub
Read BEFORE any work, in this order (normative, not optional): When work is begun via a skill or from another session, this root file may not have auto-loaded — use the start-work skill (or product-setup for first run); both self-bootstrap this reading order before any work, so the rules are never skipped. Non-negotiables (details live in the files above; on conflict, those files win): - The agent prepares, the human decides. Never invent; missing data = — to clarify —.…
In this repository2 other files it gives its agents.
- product-setupSkill
- start-workSkill
The file
We split this CLAUDE.md into the rules it gives an agent
5 rules. About 0 are common, 3 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 11
Read BEFORE any work, in this order (normative, not optional):
- Very specific: line 19
When work is begun via a skill or from another session, this root file may not have auto-loaded — use the start-work skill (or …
- Very specific: line 25
The agent prepares, the human decides: Never invent; missing data = — to clarify —.
- Distinctive: line 29
A handoff restores state — not rules, not truth: Verify its claims against the registers and artifacts.
- Distinctive: line 35
One mechanism, one way: Never introduce a second format/path for something the framework already does one way.
≈ “Common” is an estimate: at least 1 in 5 of the 28,029 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Agent rules — very-ai-product-loops
Read BEFORE any work, in this order (normative, not optional):
process/OVERVIEW.md— the model + the philosophy the agent lives by (§1)process/OPERATING-LOOP.md— how one pass of a step runsprocess/CONVENTIONS.md— notation: confidence tags, IDs, forks, change logsprocess/REGISTERS.md— register schemas- The instance: its
HANDOFF.md→sources/INDEX.md→ only the artifacts the task needs
When work is begun via a skill or from another session, this root file may not have auto-loaded —
use the start-work skill (or product-setup for first run); both self-bootstrap this
reading order before any work, so the rules are never skipped.
Non-negotiables (details live in the files above; on conflict, those files win):
- The agent prepares, the human decides. Never invent; missing data =
— to clarify —. - Registers are the home of values. Metric readings land in
registers/metrics.csvas dated rows at capture time; asources/snapshot is evidence, not the home. A data-gathering errand is still a loop pass: it ends with register updates and a change-log entry. - A handoff restores state — not rules, not truth. Verify its claims against the registers and artifacts; run its environment checks before relying on them.
- Read the tool before filling. Open
tool-skills/library/<tool>/SKILL.mdbefore writing its section. Missing prerequisites → ask or help obtain; never proceed on a guess. (Pluggable skills live undertool-skills/:library/methods ·operations/runtime skills ·adapters/output — pick by task phase; seetool-skills/README.md.) - One mechanism, one way. Never introduce a second format/path for something the framework already does one way.
- Confidence tags on every claim; agent proposals marked ⚙️. Never blanket-source your own derived conclusions.
- No secrets or PII in artifacts, handoffs, or chat. Raw captures are deleted after their values land in the registers.
- In chat with the human: no bare IDs or links — decode what each one means in the same sentence.
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.

