Weft — Go-to-Market Decision¶
Status: Decided (2026-06-05) · Owner: PM · Audience: maintainers + the in-flight docs work
One-page record of how Weft ships. Captures four decisions made 2026-06-05.
Supersedes the membership/packaging assumptions in README.md and
federation-map.md (both stale — see "Corrections" below).
Decision in one paragraph¶
Weft ships as five separately-installable products behind one brand and one front door — not a monolith-with-plugins, and not five disconnected projects. Each tool stays its own repo + its own package (PyPI / crates.io), installable and useful with zero siblings present. They are unified at the discovery and branding layer (one GitHub org, one hub, one narrative, one compatibility matrix), never at the code layer. We make a one-time coordinated Weft 1.0 launch, then every tool versions independently from that point on.
Why — two axes, not one¶
The "single product vs separate products" question is two independent decisions:
- Axis A — packaging/architecture: SETTLED by the federation axiom. Weft's differentiator is solo-useful, enrich-only, never load-bearing. A monolith-with-plugins would destroy that. Separate artifacts is a constraint, not a choice. Not relitigated.
- Axis B — distribution/discovery/branding: the real question. Today's failure mode is fragmentation (five repos, two orgs, no front door), not over-coupling. We unify the front door, not the code. "Separate packages, single front door" is the steelman of the "single product" instinct.
The four decisions¶
| # | Decision | Choice | Consequence |
|---|---|---|---|
| 1 | Org / brand | Consolidate into one weft GitHub org. |
Migrate foundryside-dev/wardline + the four tachyon-beep/* repos under one org. Set up redirects; update badges/CI/links. |
| 2 | Suite version | One-time coordinated Weft 1.0 now → independent semver + compatibility matrix forever after. | 1.0 = the four-tool code-and-debug toolkit (filigree, loomweave, wardline, legis), gated by wardline (pre-1.0) and legis (rc2→1.0). Charter is not a 1.0 gate — it joins on its own cadence. After 1.0, no monolithic release train; the "suite version" becomes a published tested-combination matrix. |
| 3 | Hero front door | Per-tool PyPI / crates.io is primary; hub is a thin index. | Invest in each tool's standalone install story (pip install filigree, cargo install loomweave, …). The hub narrates + routes; it is not the install path. |
| 4 | Claude Code marketplace | Deferred, not dropped. | A Weft plugin marketplace (mirroring the proven ~/skillpacks marketplace.json pattern; filigree already ships a filigree-workflow skill) is a post-1.0 distribution play, not a launch surface. |
Members (authoritative — kill the stale lists)¶
Weft 1.0 = a code-and-debug toolkit for agentic development — the four tools that form the build → debug → ship loop. Charter sits upstream of that loop (it defines what must be true before code) and is forward-leaning, so it ships on its own cadence and does not gate 1.0.
| Product | Lang | Owns | Launch state |
|---|---|---|---|
| filigree | Python | Work state: issues, plans, observations, findings | 2.3 — shipped · 1.0 unit |
| loomweave | Rust (+Py plugin) | Code identity + structural entity graph | 1.0 — shipped · 1.0 unit |
| wardline | Python | Trust-boundary taint analysis → SARIF | pre-1.0 — the 1.0 gate · 1.0 unit |
| legis | Python | Git/CI, governance, attestations | 1.0.0rc2 — 1.0 unit |
| charter | Python | Requirements, traceability, verification evidence | early — non-gating; joins on own cadence |
shuttle = future 6th member (transactional scoped change execution); no repo, not in scope for 1.0.
The forward roadmap (charter, then shuttle) extends Weft up into requirements and down into change execution — bookending the code-and-debug core.
Docs topology (steer for the in-flight doc work)¶
Follows directly from "separate packages, single front door":
- One Weft hub owns: the narrative, the authoritative member list (above), the federation/compatibility matrix, and the pairwise composition recipes.
- Five per-tool doc sites stay solo-complete — never assume a sibling is installed (mirrors enrich-only). filigree + wardline already have mkdocs sites.
- One source of truth for membership + contracts. Today three docs claim canon and disagree; collapse to one, vendor/transclude the rest.
Corrections this doc makes to existing artifacts¶
README.md/federation-map.mdsay "four products incl. shuttle." Wrong. Five products incl. legis + charter; shuttle is future.- The three "canonical" docs (hub
README,federation-map,~/clarion/docs/suite/weft.md) disagree on membership. This is a go-to-market defect, not a cosmetic one — a suite whose canon can't name its members can't be marketed as a suite. Fixing it is launch-blocking.
Immediate next actions (launch-gating)¶
- Publish the authoritative member list (above) and retire the shuttle-as-#4 framing everywhere. (Hand this section to the doc agent now to stop drift.)
- Reconcile the three canon docs to a single source of truth.
- Drive wardline (and legis rc2) to 1.0 — these gate the coordinated launch. Charter is explicitly not on this path.
- Stand up the
weftorg and migrate the five repos. - Publish the v1.0 compatibility matrix (which versions speak the
weftHTTP generation, which pairwise integrations are live vs asterisked).
Open risks¶
- Coordinated-1.0 gate stalls the launch if wardline slips. Mitigation: the launch is a brand moment over already-shipping tools; filigree/loomweave need not wait to keep releasing independently. Charter is descoped from the gate.
- Org migration breaks inbound links / PyPI-to-repo URLs. Mitigation: GitHub redirects + a badge/URL sweep before announcing.
- Per-tool-first front door under-sells composition. Mitigation: the hub's composition recipes carry the suite story the package pages won't.