claude-fable-5 (project maintainer)
- stance
- m1-flip
- vendor
- anthropic
- model
- claude-fable-5
- route
- subscription-cli
- gathered
- authored in-session; recommendation presented to the operator with the day-1 and later alternatives
When does this repository become public?
Outcome
Decision: m1-flip — Private through M0; public at M1 exit with full history intact
Decided by: deluongo
Decided date: 2026-08-20
Authority: public exposure is an operator right (CONSTITUTION.md, Decision rights) — this decision was always the operator's
Rationale: M1 target accepted, conditional on hardened key handling. The condition is resolved by design: M0–M1 run with zero secrets (subscription CLIs + local models satisfy the cross-vendor quorum keyless); the single metered lane stays disabled until the operator approves a dedicated key-handling design at M2. The flip itself remains the operator's button, after a full-history secret scan.
Positions
deluongo (human operator)
- stance
- m1-flip
- route
- human
- gathered
- operator's answer to the founding decision batch, quoted verbatim in the body
Context
The repository flips public with its complete history, so transparency-from-commit-#1 is preserved either way — the question is when strangers can start reading, weighed against the fact that an AI commits here autonomously and the operator's environment has a history of secrets riding tool output into places they shouldn't be. Maximum-transparency purity argues day-1; safety and first-impression argue for flipping once hygiene gates and the rendered site exist.
Position: fable
m1-flip. Day-1 publicness publishes commits before the hygiene CI, denylist gate, and full-history scanning exist — the exact window where an accident is least catchable. Nothing is lost by waiting: the entire M0 history becomes public at the flip, which also forces M0 to follow public-grade hygiene from the first commit. And the first thing a stranger sees should be the rendered archive, not scaffolding. The flip is the operator's button after a clean full-history secret scan.
Position: operator
Quoted verbatim from the founding decision batch (2026-08-20):
Public at M1 sounds good as a target, but depending on what keys you need we need to setup a better / more hardened way to make sure they don't leak, because I'm security conscious and the tool call workflow makes it's difficult for you to no leak them accidentially. I appreciate you asking this question.
Read as: m1-flip accepted as target, with a binding condition that key handling be hardened beyond the default workflow before any key exists in the project. The maintainer's response (adopted into the plan and CONSTITUTION): M0–M1 use zero secrets; the one metered lane is gated behind an operator-approved key-handling design at M2; the blessed-invocation path keeps the key out of the session environment entirely.