plainweave
Enables tracing code entities through requirements to strategic goals, surfacing orphan code and allowing query of intent via graph primitives like orphans, trace, and corpus.
README
Plainweave
Permission for code to exist.
Plainweave is the Weft federation member that holds the team's code-grounded intent: a traceability graph in which every code entity must earn its existence by laddering up to a requirement, and every requirement must ladder up to a strategic goal. A node with no upward edge — at any level — is a reviewable question:
- a public capability with no requirement → "why does this code exist?"
- a requirement with no goal → "what am I doing here?"
The point is not catching orphan code (that is one query). The point is that binding code to requirements accretes a living, readable requirements corpus an agent or human can reason over — "why do we have three requirements for reporting that are all the same?" — and consolidate. Plainweave moves the refactor lever up the Meadows leverage hierarchy: from the lowest altitude (rename this function, extract that class) to a high one (why does this submodule exist; does it still serve a goal we hold?).
Status — 1.0. The build is complete and stable: the intent graph, the read primitives, the authoring surface, the cross-member seams, and
doctorship with versioned JSON envelopes, strict typing, and a green gate (lint, mypy strict, tests at ≥90% coverage). This is stable behaviour and contracts — cross-language coverage completeness (e.g. Rust public-surface tagging upstream) is a documented roadmap item, not a 1.0 gate. Formal Weft suite membership / launch-cutover inclusion remains owner-gated. Reframed and renamed from the~/charterprecursor; the backlog lives in this repo's.filigreetracker.
The model — a traceability graph of intent
strategic goal ──▲── requirement ──▲── code SEI (leaf)
(root intent) (obligation) (the thing that exists)
- Leaves are code entities — Loomweave SEIs (modules + public surfaces). Interior nodes are typed intent nodes (requirement, strategic goal) at any altitude; altitudes are just node types, the graph does not fix the number of levels.
- Edges mean "justified by / satisfies."
- Requirements are trivially mintable (shells welcome). The corpus tolerates mess by design — value comes from the mess being visible and queryable, then consolidated. Cheap minting feeds the corpus.
- Code leaves are keyed by Loomweave SEI, so bindings survive rename/move.
- Default trace altitude: modules and public/exported surfaces must trace; private internals inherit their container's justification.
A thin member — Plainweave builds none of its siblings' machinery
Plainweave is advisory by default and deliberately thin on teeth and audit. It owns the intent graph and the reasoning reads; it delegates everything else.
| Tool | Owns | Plainweave does not rebuild |
|---|---|---|
| Plainweave | the intent graph (goals ↔ requirements ↔ SEI bindings) + the reasoning reads. Its domain: accreted, code-grounded intent. | — |
| Loomweave | the entity catalog (what exists; public vs internal), SEI identity, the rename feed, and the semantic-search engine. | identity/rename tracking; embeddings |
| Legis | the git/CI boundary surfacing, all graded enforcement (advisory default; dial-up per repo via policy cells), and the audit trail. | enforcement engine; override/audit |
Bindings reuse the ADR-029 entity-association contract (SEI-keyed, with
content_hash_at_attach drift detection — the same pattern Filigree uses to
bind issues to code), not a new link store.
Surfaces
Write path — authoring-time binding ("speak SEI at entry," extended to intent). When an agent creates or commits a module / public entity, Plainweave offers an inline bind: link this SEI to a requirement (existing or a freshly minted shell) and optionally ladder that requirement to a goal. Cheap, inline, attributed. Code that skips the bind is exactly what surfaces as an orphan.
Read surfaces — three composable graph primitives (built for unanticipated agent use, not canned reports):
orphans(level)— unlinked nodes at the code / requirement / goal altitude.trace(node)— up to goals, down to code.corpus()— the readable dump of requirements with their code- and goal-links: the artifact a curator reads to spot "these three are the same." Consolidation is agent-driven; Plainweave serves the substrate, not an automated verdict.
Over these three sits coverage() — the self-computed intent-coverage
north-star: the fraction of public surfaces that answer "why does this exist?"
It is honestly qualified in-band (namespace scoping, denominator_complete,
present_plugins, bounded evidence) and never reports a silent clean when the
denominator is partial.
Boundary — coverage facts ride out at the git/CI boundary through Legis ("this change adds N public entities bound to no requirement"). Advisory by default; any repo wanting teeth dials it up through Legis's policy cells. Plainweave adds no enforcement of its own.
Optional similarity hint — Loomweave now ships semantic search, so a thin "these requirements look like the same thing" hint becomes reuse of a proven sibling capability rather than a from-scratch ML build. It assists the curator; it is explicitly not a dedup engine.
Doctrine fit
- Coordinate, not gate — advisory default.
- Enrich-only — Plainweave absent → Loomweave, Legis, and the code are unaffected; solo mode degrades to manual file/symbol refs.
- Speak-SEI-at-entry — binding at authoring keeps code on the moat.
- Don't-duplicate — Legis owns teeth + audit; Loomweave owns identity + semantics.
- Prescribe-nothing — a general graph + queries; agents compose uses we haven't imagined.
Cross-member seams
The seams (Plainweave → Loomweave catalog/rename/semantic; Plainweave → Legis boundary) are hub-blessed and prove-the-need: built as additive adapters on Plainweave's side, never pre-frozen sibling obligations until the need is shown live (golden vector / live consumption). Each seam ships with a blast-radius map + dated counterpart tickets.
Documentation
docs/design/2026-06-18-plainweave-permission-to-exist.md— the canonical design ("permission for code to exist"). Start here.docs/MODULE-MAP.md— what the precursor core carries forward vs. what the reframe reshapes.docs/README.md— index of canon vs. precursor-era docs.
Installation
pip install plainweave
Or with uv:
uv pip install plainweave # add to an environment
uvx plainweave --help # or run it without installing
Plainweave requires Python ≥ 3.12. Installing exposes two console commands:
plainweave— the CLI:init,intent(coverage/orphans/trace/corpus),req,goal,bind,catalog,criterion,verify,status,dossier,baseline,actor, anddoctor.plainweave-mcp— the read-only MCP server that mirrors the intent reads for agents (mutates:false,local_only:true).
Quick start:
plainweave init # create a local store under .plainweave/
plainweave intent coverage # the north-star: how much public surface is justified
plainweave doctor # check store, Loomweave binding, and MCP surface
Development
Plainweave uses uv, hatchling, ruff, mypy,
and pytest.
uv sync --group dev
make ci # lint + typecheck + test (coverage-gated)
The runtime package depends on the official Python MCP SDK for plainweave-mcp.
Web UI (optional)
Install the extra and launch the operator console:
pip install 'plainweave[web]'
plainweave web --actor human:<you>
Browse the corpus, author requirements, and ratify agent-proposed drafts and trace links. Local-first, single-operator; advisory only (no release verdicts).
Accessibility (AT gate — manual)
Before shipping the review surface, run an NVDA (Windows) or VoiceOver (macOS)
pass over the /review queue. Each approve / accept / reject action must:
- Announce the outcome via the
#sr-statuslive region (role="status" aria-live="polite"), e.g. "Approved: Requirement title". - Move focus to the next action button in the queue (or to the "All caught up" heading when the queue empties).
- On the last item, announce "Queue is now empty" and place focus on the "All caught up" heading.
Structural contracts (live region presence, skip-link, labelled search input,
per-item aria-label on Approve buttons) are locked by
tests/web/test_a11y_contracts.py. The focus-management and announcement
behaviour above require a live AT session and cannot be automated in the
current test harness.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.