CLI & Scripts Reference
Every wf command, every page type, and the module layout of the harness. For configuration (providers, routing), see Configuration.
Note Types
Every page in the fabric is one of 24 types (VALID_TYPES in lint.py — the corpus has grown; the lint code is the source of truth). The type drives what lint demands of it and who can change its status — that's how a claim and a pattern can have different trust rules. The ones you'll touch daily are claim, experience-event, pattern, and decision.
| Type | Location | Purpose |
|---|---|---|
source | evidence/sources/ | Immutable bibliographic record + sha256 |
| └ capture kinds | evidence/raw/<slug>/git/, evidence/raw/<slug>/chats/ | kind:-discriminated capture shapes (#156): pr-record (github route — source_repo/pr/pr_state/merged_at, thread-graph node), commit (local route — sha + date + message, metadata only), chat-session (session/harness/session_started/files_touched/related_sessions). Contracts: schemas/frontmatter.md §source |
source-summary | evidence/source-summaries/ | Faithful summary with locators, no inference |
claim | evidence/claims/ | Atomic assertion with source_refs (locator + quote) + code_symbols + graph_edges |
concept | concepts/ (unbound) or domains/<domain>/concepts/ (bound; canonical via the ontology alias map) | Stable explanation built ONLY from linked claims |
experience-event | projects/<p>/experience-events/ | Structured observation: problem → intervention → outcome |
pattern | patterns/ | Reusable context→problem→forces→solution (maturity 0–3) |
anti-pattern | anti-patterns/ | Repeated failure mode; detect and warn |
skill | skills/ | Reusable procedure with inputs/outputs |
decision | projects/<p>/decisions/ | ADR-like technical choice record |
entity | global/entities/ | Code-symbol index from AST parsing |
change-set | evidence/traces/change-sets/ | Agent-proposed edit batch for human review |
synthesis | syntheses/ | Query answer filed for reuse |
See schemas/frontmatter.md for full contracts.
Lifecycle: how pages move through statuses
Types carry a status vocabulary, and transitions are gated — some by lint, some by humans. The two lifecycles you interact with daily:
Claims (evidence lifecycle):
text
proposed ──(locator verified)──> supported ──(newer claim wins)──> superseded
│ │
└──(evidence contradicts)──> contested retracted (bad quote/source)proposed: extracted but unverified (nosource_refsrequired yet)supported: locator-verified, quote verbatim — lint demands source_refscontested: contradicted by another claim — both stay visible withcontradictsrelations; a human settles it, a model never averages itsuperseded: requiressuperseded_by: [[claim-...]]— chains can't loop
Patterns (maturity ladder):
text
candidate (m0-m1) ──(observed in ≥2 independent projects)──> recommended (m2) ──(human review)──> standard (m3)
│ │
└────────────────────── deprecated ←──────────── any time ─────────────────────────────┘recommendedis lint-enforced to have maturity ≥ 2 and evidence from ≥2 independent projects (independent = the observations came from different codebases, not one project copied to another (the independence rule))standardrequires ahuman:verifier — the machine can never bless a pattern into load-bearing statusdeprecatedkeeps the page for history; context compilation stops selecting it
Who does what: agents propose (claims, change-sets, dossiers — always staged, never merged silently); humans promote (merge change-sets, approve dossiers, set standard). The gates are where governance lives.
CLI Reference (wf)
The wf command is the single entry point for controlling the fabric. It installs to ~/.local/bin/wf (alias wiki-fabric).
| Command | Purpose |
|---|---|
uv tool install wiki-fabric | Install the tool (one line, atomic — the packaged mode; --with mcp adds the MCP server) |
wf install [--dir DIR] [--with-graphify] | Create the fabric content skeleton + starter fabric.yaml at WIKI_FABRIC_DIR (or cwd); the corpus is content — sync it with wf sync setup/init |
wf update | Dev mode: pull the harness + rebuild hooks/index/catalog. Packaged mode: hints uv tool upgrade wiki-fabric (content updates are wf sync pull, not tool updates) |
wf status | Fabric health + inventory counts |
wf doctor [--json] | Environment diagnosis: endpoint reachability, model resolution (phantom compiler_model), compiler-eval readiness, judgment-tier availability, vault drift, gate — each failure names its fix |
wf completions {bash|zsh|fish} | Emit a shell completion script on stdout — verbs + subcommands (mine chats|promotions, eval behavior|…, sync …) + common flags per verb. Install: eval "$(wf completions bash)" (bash); wf completions zsh > "${fpath[1]}/_wf" && compinit (zsh); wf completions fish > ~/.config/fish/completions/wf.fish (fish). Static generation — never breaks offline |
wf status --json | The inventory/provenance surface as JSON for agents/hooks: fabric/vault paths, CLI mode, LLM config, inventory counts + connected repos, per-repo capture provenance (last capture, window, git/chats counts), lint health, gate sections, graphify state |
wf projects [--json] | Connected repos one-per-block: slug, path, owner, extract route, capture counts (raw/git/chats/claims), last git capture + window; --json for agents |
wf overlay-track [--dry-run] | Track every connected project's .wiki-overlay.md in its repo (H4 sweep): stages + commits untracked overlays — a second machine's clone then carries the config. Idempotent; tracked repos skip. Lint (OVERLAY-UNTRACKED) + doctor name anything left |
wf vault [PATH] | Scaffold/audit the Obsidian output vault |
wf update | Tool upgrade + auxiliary refresh (hooks reinstall from config, entity index, catalog) — packaged mode prints uv tool upgrade wiki-fabric; see Upgrading wf |
wf bootstrap <project-path> | Connect a project to the fabric |
wf capture <project-slug> [--repo PATH] | Capture upstream repo docs → evidence/raw/. Exit codes: 0 nothing new, 2 drift captured, 3 unknown project slug (with the connected-project list — hooks/CI can detect a typo) |
wf capture <project-slug> --git <owner/name-or-path> | Capture PR/issue threads + high-signal commits → evidence/raw/<slug>/git/ (--since 6m window, --limit 30 BUDGET — the window auto-shrinks to fit an active repo, never truncates; --until backfill bound; threads paginate + flag truncation, the count persists in .last-capture as truncated=N and surfaces in wf status; --churn; --since-state incremental hooks; knobs: tuning.git_history + repos.<slug>.git_history). Unknown slug exits 3 (the #167 table). --until does not compose with --since-state — separate flows by design: --since-state means "incremental from the recorded base" (the recurring path), --until means "one bounded backfill slice" (an explicit, opt-in window override); composing them would put two bounds on the same axis. Never captured + want the slice? Run one --until slice first, then --since-state picks up from there |
wf capture chat <project-slug> | Capture agent chat sessions → evidence/raw/<slug>/chats/ (harnesses: claude, opencode, codex, gemini — auto-detects; --since 90d, --min-turns, --harness <name>) |
wf capture issues <project-slug> --tracker github --repo owner/name | Capture tracker issues + comment threads → evidence/raw/<slug>/issues/ (#176; kind: issue-record thread nodes): --since 6m updatedAt window + budget (shrink-not-truncate), --since-state incremental (own state stamp <slug>-issues), --until backfill slice (separate flow), --no-comments; deterministic density pre-filter (threads/rationale/acceptance-criteria kept, chores skipped). The WHY channel: PRs carry what, chats how, issues why (spike: 44 claims/3 issues, ~7% metadata noise) |
wf context --task "<task>" [--project <slug>] [--paths P] [--write-receipt] [--format json] [--judge-borderline] | Compile the task-scoped context manifest (0 tokens; --write-receipt persists a delivery receipt). --judge-borderline: opt-in judged re-rank of borderline beyond---max candidates (needs integrations.judgment; per-item receipt record) |
wf ingest <source> [--extract-claims] | Ingest a source (LLM claim extraction; claims carry provenance edges from chat/PR captures) |
wf ingest --changed <slug> [--budget N] | Ingest only NEW/CHANGED raw files for a project (sha256 anti-loop; --budget/tuning.ingest.budget caps the run — re-run continues). Drift before ingest: old-revision claims mechanically flip contested+stale_after (0 tokens) |
wf ingest --pending <slug> [--budget N] | Claim-extract recorded-but-unextracted sources (anti-loop safe; budgeted like --changed) |
wf ingest --reclaim <slug> | Recover zero-claim ingested sources → back to pending (silent-orphan recovery, #93) |
wf query "<question>" [--no-rerank] [--type changed] | Ask the fabric a question (lexical + graph expansion; lineage-shaped queries gain an evidence-graph provenance section; graphify symbol discovery when enabled; optional System One fusion rerank — a local decision model reorders the top candidates, latency-budgeted, silent fall-back to lexical order). #189: --type changed (auto-detected for "what changed / how did / evolution" queries) orders the answer earliest→latest by capture date with per-step locators — the temporal axis |
wf freshness [--dry-run] [projects...] | Scheduled upstream-freshness cycle (0 tokens): capture-git --since-state + mechanical auto-reverify per connected repo — including the newest-wins contradiction sweep (#175; a demotion is gate-worthy drift); scheduled on the corpus CI by sync setup/init (opt-in var WIKI_FABRIC_FRESHNESS=1); drift flags land in wf gate. Exit: 0 clean, 1 drift recorded, 2 hard failure (no fabric / unknown slug — the connected list names the fix) |
wf thread <session-or-claim-id> [--stats] [--json] | Evidence-graph thread lookup: claims citing the session/PR, files touched, continuation edges |
wf log --project <slug> ... | Log an experience event |
wf remember <project> --note "..." [--kind feedback|project|user|reference] | Friction-free memory note (#177 S6 — the auto-memory lesson): one line, 0 tokens, no LLM, no claim ceremony. Mining reads standing notes (a repeated note becomes a promotion candidate); ephemeral - auto-expiry (feedback 45d / project 30d / user 90d / reference 90d; --expire sweeps, --dry-run reports). --list shows standing notes |
wf hook {install|uninstall|status|reinstall} | Git post-commit auto-capture+ingest (--extract-claims for LLM on drift). The commit body (v6) also runs the commit-time LAYOUT-GUARD on the harness repo's own scripts — a corpus-path re-spell fails the commit with the fix pointer, not at the next test run |
wf claude legacy | Pre-harness always-on installer (always_on.py; superseded by wf harness install) |
wf okf export --out DIR [--scope S] | Export the fabric as a deterministic portable OKF v0.2 bundle |
wf okf import <bundle> [--scope S] | Ingest an external OKF bundle as immutable evidence (trust recorded, not inherited) |
wf sync {setup|init|status|push|pull|resolve|commit-drift} | Share the corpus with a team — setup uses the gh CLI to create + publish the corpus repo; resolve without --strategy runs the interactive resolver (diff + pick ours/theirs/union/skip). Team mode (sync.mode: team) opens one PR per push — evidence auto-merges on green CI, atoms wait for human review; sync push --pr opts in per-invocation; commit-drift stages + commits hook-accumulated drift with a plane-classified summary (never pushes); init/setup scaffold promotion-queue.md + questions/ |
wf skill [--list] [<name>] | Print the procedure for a workflow (ingest, promote) — universal across all agent harnesses |
wf harness {install|status} [--all|--only k1,k2] [--force] | Install always-on + skills into detected agent harnesses (11 supported; wf claude is the legacy alias) |
wf models ensure [--model ID] [--yes] | Check llm.local_model is cached; offer human-gated download (--check exits 0/1 without prompting) |
wf review --check [--project <slug>] | Staleness report: current, due for review, overdue, stale |
wf review --verify <claim-id> | Re-verify a claim (stamps last_verified, rolls review_after forward by tier) |
wf review --auto-reverify | Mechanically re-verify all overdue claims (sha256 + quote check, 0 tokens) |
wf review --verify-sources [--dry-run] | Mechanically drive the SOURCE lifecycle (#160): sha256 recompute per record — fresh rolls review_after (capture-kind tier: pr/commit 30d, chats 45d, docs 180d), upstream-drifted stamps stale_after, upstream-DELETED expires the record (status: expired — provenance tombstone, excluded from context) (0 tokens) |
wf review --contradiction-sweep [--dry-run] [|--json] | Newest-wins demotion (#175): a claim contradicted by a STRICTLY NEWER still-supported claim mechanically flips contested + stale_after (reason line records the contradictor). Carriers: hand-written contradicts relations + registry/effects/ judged verdicts. Restore is automatic when the contradictor loses support (or --restore-demoted <claim> by hand); --demoted lists the demoted set. Equal clocks never demote (intra-batch reads stay human-owned). Exit 1 when a demotion landed (gate-worthy) |
wf review --verify-locators | Re-check every claim's locator against its raw source: rewrite drifted locators, repair backtick-elision quotes, restore wrongly-contested claims, strip stamps + contest vanished quotes (0 tokens, #109) |
wf gate [--quiet|--json|--write-manifest|--deliveries] | Aggregate every pending human decision (overdue/stale claims, promotion dossiers, domain proposals, pattern candidates, open questions, + knowledge sources going dark — projects with no contribution in 30d, info tier) into one report. Exit 1 when anything is actionable. --write-manifest persists registry/pending-gate.md; --deliveries surfaces recent context receipts (was the manifest right?) |
wf relocate-concepts [--dry-run] | Move domain-bound concepts to their physical homes domains/<domain>/concepts/ (S1 migration; deterministic, idempotent, alias-folded, 0 tokens) |
wf promote-domains {list|--apply <dossier>} | Human-gated merge of an approved domain proposal into domains/ontology.md |
wf propose-domains [--dry-run] | Propose new domains from corpus clusters (0 tokens; staged for human review) |
wf harvest-questions [--project <slug>] [--dry-run] | Harvest concept open-questions → staged question pages (priority from claim confidence, 0 tokens) |
wf promote-questions {--list|--open|--apply <id>|--reject <id> --reason} | Human-gated apply/reject of harvested questions (--open lists canonical open questions) |
wf verify-effects <claim.md>... [--source <slug>] [--max-pairs N] | Judged effect verification at ingest (independent second opinion; writes <claim>.effects.json) |
wf wiki-generate {begin|next|submit|finish|status|inspect} | Wiki-generation bookkeeper (writer/bookkeeper split): durable page queue, sparse claim deltas, citation validation, run checkpoints |
wf publish [--out DIR] | Publish the wiki as a static site (Quartz v4, pinned revision; graph view + search) |
wf utility [--dry-run|--json] | Receipt↔outcome join → usage counts on pattern pages (0 tokens) |
wf integrations | Show optional-integration state (graphify, obsidian, judgment, embeddings) + what each changes |
wf export wiki [--mode mechanical|llm|hybrid] [--project <slug>] [--full] [--check] | Generate the human-layer wiki: topic articles, project retrospectives, staleness dashboard. Writes OpenWiki-style pages (SUMMARY lead, Key Takeaways, Sources backtrace, provenance stamp), validates/repairs Mermaid diagrams, and emits the citation graph (registry/wiki-graph.json). #187 delta-mode: unchanged generation-inputs keep the previous file byte-identical (0 tokens); changed inputs patch incrementally with mechanically-spliced untouched sections (--full forces wholesale; --check reports unchanged/stale/never, 0 tokens). Browse the [[wikilinks]] in Obsidian's native Graph view. |
wf export-viz [--out DIR] | Export the static wiki-graph viewer (#177 S5): self-contained index.html + graph.json (the citation graph the export builds) - static-host ready, no live reload |
wf mine chats <project> [--llm] [--propose] [--dry-run] | Distill captured chat transcripts into durable takeaways (transients filtered); --propose stages pattern/anti-pattern candidates in patterns/_inbox/ (gated, provenance-cited, idempotent) |
wf mine promotions [--judge] [--min-projects N] | Cluster experience events → promotion dossiers (deterministic, 0 tokens; --judge for the near-miss judgment tier) |
wf promote-patterns {--list|--apply <id>|--reject <id> --reason|--auto|--unapply <id>} | Human-gated apply/reject of chat-mined pattern candidates; --auto is the #190 judged-confident tier (opt-in tuning.promotion.auto_apply, G-J gated, band-banded: p ≥ auto_threshold applies + stamped + reversible, near-band escalates to the gate, below stays annotated); --unapply rolls an auto-apply back |
wf version | Show wf version + CLI sync state (installed ~/.local/bin/wf vs harness script) |
wf --version / wf -v | Same version one-liner as a flag (packaged mode: version + harness root) |
wf help <command> | Per-verb help — routes to the underlying script's argparse (wf help ingest = wf ingest --help); unknown names get nearest-verb suggestions |
wf repos migrate [--dry-run|--apply|--prune] | Move per-repo routing from fabric.yaml into project overlays |
wf lint [--okf] [--format json] | Deterministic linter (full profile; --okf = OKF conformance floor; JSON for CI) |
MCP Server
wf-mcp (installed with uv tool install wiki-fabric --with mcp) exposes the fabric's core surface as native MCP tools for any MCP client:
| Tool | Wraps | Cost |
|---|---|---|
fabric_query | wf query | 0 tokens |
fabric_context | wf context | 0 tokens |
fabric_gate | wf gate | 0 tokens |
fabric_thread | wf thread | 0 tokens |
fabric_log | wf log | the mining intake |
wiki_begin | wf wiki-generate begin | 0 tokens (deterministic outline) |
wiki_next | wf wiki-generate next | 0 tokens (next page job) |
wiki_submit_page | wf wiki-generate submit | 0 tokens (validates + persists) |
wiki_finish / wiki_status / wiki_inspect_page_claims | wf wiki-generate … | 0 tokens |
Mutations beyond log used to stay CLI-gated; the wiki-generation tools are the one deliberate extension — the writer/bookkeeper split (#144): the host
Exit-code contracts
Hook/CI shell-outs key off these — the table is the contract (don't renumber without reading how consumers parse, test_shim_parity-style, first):
| Verb / script | Code | Meaning |
|---|---|---|
wf capture | 0 | nothing new |
| 2 | drift captured (hooks may trigger ingest) | |
| 3 | unknown project slug (+ connected list, closest-match hint) | |
| (1) | other errors (repo path missing) | |
wf capture --git (capture-git) | 0 | nothing new |
| 2 | drift captured | |
| 3 | unknown project slug (+ connected list) | |
wf capture issues | 0 | nothing new |
| 2 | drift captured | |
| 3 | unknown project slug (+ connected list) | |
wf freshness | 0 | clean (no drift) |
| 1 | drift found + recorded (gate-worthy) | |
| 2 | hard failure (no fabric, unknown slug) | |
wf gate | 0 | nothing pending |
| 1 | actionable items want a human | |
wf models ensure --check | 0/1 | model cached / needs download (script-friendly, no prompt) |
wf review --verify-sources | 0 | source lifecycle driven (refresh/stamp/expire counts reported) |
wf review --contradiction-sweep | 0 | no demotion |
| 1 | a demotion landed (gate-worthy) | |
| agent writes prose, the MCP tool validates citations and reconciles claim | ||
deltas deterministically (the bookkeeper makes no model calls). Claude Code: claude mcp add wf-mcp wf-mcp; Claude Desktop/Cursor: stdio-server config pointing at wf-mcp. |
Environment: WIKI_FABRIC_REPO overrides the source repo URL.
Scripts Reference
Scripts live in four subdirectories of scripts/ — cmd/ (entrypoints), eval/, harness/, and shared lib/. Each runs standalone (<name>.py --help) — but the wf verbs are the user surface: wf eval {behavior|stability|golden|real|pr}, wf graphify {all|import|enrich|diff|status} wrap the eval/bridge scripts.
cmd/ — CLI commands
| Script | Purpose | LLM Cost |
|---|---|---|
capture.py | Capture upstream repo docs → evidence/raw/ | 0 |
capture-git.py | Capture git history (PR/issue threads, commits) | 0 |
capture-chat.py | Capture agent-harness chat sessions as chat-transcript evidence | 0 |
ingest.py | Source → source record → claims (LLM extraction) | 1 call per source |
query.py | Question → evidence-backed answer | 0 tokens (lexical + graph) |
context.py | Task → scoped context manifest with reasons (--write-receipt persists a receipt-v1 artifact) | 0 tokens |
lint.py | Deterministic checks (incl. --okf floor, llm.local_model) | 0 |
synthesize.py | Claims → concept pages (cloud or on-device route) | 1 per concept |
log-experience.py | Capture experience event (wf log) | 0 |
mine-promotions.py | Cluster experience events → dossiers (wf mine promotions) | 0 |
promote.py | Manage pattern promotion dossiers (wf promote) | 0 |
propose-domains.py | Discover + propose new domains (pending-review dossiers) | 0 |
promote-domains.py | Human-gated merge of a proposed domain into the ontology | 0 |
gate.py | Aggregate pending HITL decisions + notify adapters (wf gate) | +--write-manifest/notify |
rebuild-index.py | Rebuild registry/catalog.json from files (wf rebuild-index) | 0 |
relocate-concepts.py | Domain-bound concepts → domains/<domain>/concepts/ (S1 migration; deterministic, idempotent) | 0 |
build-entity-index.py | AST entity index from source repos | 0 |
bootstrap-project.py | Connect a new project to the fabric (wf bootstrap) | 0 |
ensure-local-model.py | Check/download llm.local_model (wf models ensure) | 0 |
review.py | Staleness scan + re-verify loop (wf review) | 0 |
export-wiki.py | Human-layer wiki renderer (topics, projects, staleness, deep dives) | 0–1 per topic |
wiki_generate.py | Wiki-generation bookkeeper — durable queue, claim deltas, citation validation (wf wiki-generate) | 0 |
verify-effects.py | Judged effect verification at ingest (wf verify-effects) | n judgment calls |
publish-wiki.py | Static-site publisher (Quartz v4, wf publish) | 0 |
utility.py | Receipt↔outcome join → usage counts (wf utility) | 0 |
mine-chats.py | Distill chat transcripts into durable takeaways | 0 or 1/session |
repos-migrate.py | Move per-repo routing from fabric.yaml into overlays | 0 |
okf_export.py / okf_import.py | OKF bundle export / import (wf okf) | 0 |
harness/ — agent-integration
| Script | Purpose | LLM Cost |
|---|---|---|
harnesses.py | Multi-harness registry + installer (11 agent tools) | 0 |
skill.py | Universal skill loader (prints procedures) | 0 |
hooks.py | Git post-commit/post-merge capture+ingest hooks (code commits also run the graphify cycle when enabled) | 0 |
always_on.py | Legacy always-on installer (superseded by wf harness install) | 0 |
graphify-bridge.py | Optional graphify integration (update/import/enrich/diff) | 0 |
eval/ — evaluation family
eval.py, eval-behavior.py, eval-stability.py, eval-pr-replay.py, eval-real-repo.py — golden corpus, behavior, stability, PR-replay, and real-repo evaluations.
lib/ — shared modules (imported, not entrypoints)
| Module | Purpose |
|---|---|
fabric_config.py | fabric.yaml loading, stage routing, actor conventions, local model, canonical repo keys (#158) |
layout.py | Corpus layout single truth: SEGMENTS + accessors + the claim-prefix join contract; seg() raises on undeclared names (re-spelling is the drift class) |
ontology.py | Ontology vocabulary single truth: ## Domains/## Aliases/## Shared tag set → canonicalize() (alias-fold; unknown binds nothing), all_spellings() |
extract_backends.py | Claim-extraction layer: prompt building, 4 LLM backends, JSON repair |
local_llm.py | On-device generation (GGUF/MLX), serialized model cache |
wf_common.py | Shared helpers: parse_frontmatter, norm, slugify, project_slug (THE canonical underscore→kebab fold), hashing |
eval_core.py | Eval scoring primitives: concept_match, jaccard, fuzzy_coverage |
judgment.py | Judgment tier backends (Jev cloud / Laya-MLX / upstream laya / generic), typed questions, graceful degradation |
embed_index.py | Hash-gated offline embedding index (registry/embed-index.json) + query-side embedding |
tombstones.py | Rejected-proposal buffer (SkillOpt S3): tombstones + cluster matching for the miner |
deepdives.py | Graphify-rendered project deep dives (architecture/components/tour, 0 tokens) |
obsidian_bridge.py | Two-way vault bridge: harvest/push/export manifest |
Shell scripts remain at scripts/ root only where shell is the right tool — wiki-fabric.sh (the curl-piped installer + dev dispatch), demo.sh, and smoke-test.sh. The change-set merge (wf apply-changeset) and vault scaffold are python now — cross-platform, testable in pytest.
Fabric Inventory
Counts live in your fabric, not the repo. Check yours anytime:
bash
wf statusThe repo ships only the harness — templates, examples (templates/examples/), schemas, and scripts. Your claims, captures, and patterns accumulate locally as you connect projects.
Vault & status commands (the human views)
wf vault [PATH] [--check] — the vault is standalone OUTPUT
The vault is the result of the utility, never a mirror: wiki-fabric reads the corpus and writes generated content (the rendered wiki) into it. It never copies or symlinks source content (evidence/, claims/, patterns/, projects/, registry/) into the vault — those live only in the fabric. Because the vault is standalone output, you can point vault.path at several targets to get multiple vaults on one machine (e.g. work vs personal).
wf vault scaffolds the output directory + .obsidian/ workspace and audits it: it flags a missing generated wiki, and flags stale leftovers from the older mirror model (corpus dirs or symlinks that don't belong in an output vault) — it does not delete them. The actual wiki pages are generated by wf export wiki into the vault. wf vault --check exits 1 on drift (CI-friendly).
wf status shows the vault verdict (fresh / structure drift — run: wf vault) inline, so staleness is one glance away.
wf repos migrate — config-per-project migration
Moves explicit per-repo routing keys from fabric.yaml into each project's .wiki-overlay.md (--dry-run to preview, --apply to write). fabric.yaml stays untouched — prune by hand after verifying. New projects don't need it: wf bootstrap --extract local writes routing directly.
wf status — the health dashboard
One screen answering "is my fabric healthy": fabric + vault locations, the three models (ops / compiler / local, with the local model's cache state), inventory counts (claims, sources, concepts, patterns, projects, entities), lint verdict, graphify import state.
wf integrations — what's active
Lists optional integrations with their state and effect:
text
✓ graphify: ENABLED (call-graph staleness, claim enrichment, graph expansion)
commands: graphify-bridge.py --all | --diff | --statusInactive integrations print their enable command and what they'd change — so the dashboard doubles as documentation of the delta (full table in Integrations).