fluxrecord v0.2.0

The ledger, the CLI and the specification are now in agreement, and the adapters
the CLI promised have been written.

Added

Wordmark — the mark is one idea in five styles: a flux waveform riding
above a half-block letterform. flux art prints it, with an optional ANSI
gradient that honours NO_COLOR.
Adapters — fluxrecord.tools (local executable inventory),
fluxrecord.dashboard (bounded, GET-only local preview) and
fluxrecord.github_sync with fluxrecord.github_cycle (durable GitHub
observations and an idempotent outbox).
Authentication — fluxrecord.auth: JWT issuance and verification,
bearer middleware, and the login, refresh, revoke, introspect and logout
routes. HS256 only, stdlib only, framework free.
Ticket files — flux export writes the layout the specification
describes: markdown tickets with front matter, an index, and one evidence
file per checkpoint.
Import — flux import reads a CSV backlog, mapping the file column onto
scope so overlapping work is detectable. Rows whose id is already used by a
different ticket are reported as conflicts rather than dropped.
Commands — flux list, flux show, flux version, flux renew,
flux reconcile, flux art and flux import.
Ticket model — the full schema from the specification: title, priority,
tags, dependencies, blocks, owner, seed key and updated time.
Migrations — a meta table records the schema version; migrations are
additive and idempotent.
Property tests — for the token rules, the token budget, the stagnation
rule, the byte budget and scope conflicts.

The documented interface now exists: flux task-add --budget and
flux decide <id> <decision> are accepted, and the acceptance gate no
longer contradicts the specification.
Work whose dependencies are not met is refused rather than silently queued.

pip install fluxrecord

Full details in the changelog.