* feat(drift): UNVERIFIED_AC multi-framework test_ref↔testcase matching (closes #203) (#204)

* feat(drift): UNVERIFIED_AC — verify test_refs ran and passed via JUnit XML (F-96700032)

UNTESTED_AC only checks that a done AC's test_refs EXIST on disk — an empty
file, a test.skip, or a failing test all satisfied the gate, so the chain
stopped at "AC → named file", not "AC → observed pass". UNVERIFIED_AC closes
that gap when a JUnit XML report is available.

- New detector UNVERIFIED_AC (src/stages/detectors/unverified-ac.ts) + a pure
  regex JUnit parser (src/stages/junit-report.ts, no XML dependency, mirroring
  the coverage-XML approach). Report path comes from gate.test_report in
  .cladding/config.yaml or a conventional default; failing/errored or
  only-skipped test_refs are an error, absent ones a warn (--strict promotes).
- Graceful by default: no report present → emits nothing, leaving UNTESTED_AC's
  existence check as the baseline, so non-JUnit projects are unaffected.
- Registered in detectors/index.ts (count auto-recomputed 37→38); catalog +
  self-consistency detector-count claims bumped to 38.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* feat(drift): UNVERIFIED_AC multi-framework test_ref↔testcase matching (F-d980359c)

UNVERIFIED_AC's matcher was effectively vitest-only — it keyed every testcase
by `classname` and assumed that was a file path. pytest (`tests.test_foo`),
Java/Kotlin FQCN (`com.example.FooTest`), and `file=`-attribute emitters never
matched, so a passing test read as `absent` (a false positive under --strict)
and a real fail/skip was mis-reported as "did not run".

- parse: index each testcase under every path-shaped key it yields — the
  `file=` attribute, the `classname` as-is, and a dot→slash conversion of a
  dotted (no-separator) classname — sharing one accumulator (no double count).
- lookup: extension-agnostic exact/suffix match (`FooTest` ↔ `FooTest.kt`).
- detector: confident-or-degrade — a report with no path-like keys (e.g. jest
  describe-title classnames) is unmappable, so emit nothing instead of flooding
  false `absent` findings (preserves the low-false-positive contract).

Measured A/B (OLD = classname-only): correct verdicts across a
vitest/pytest/Kotlin/jest matrix 2/8 → 8/8; parse cost on a 10k-case report
+1.6ms (vitest) … +7.8ms (pytest) per gate run — noise vs the ~50s gate.

Blind-authored tests (tests/stages/junit-multiframework.test.ts) cover all 5
ACs. Existing UNVERIFIED_AC tests stay green (backward compatible).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

* feat(ears): add the 6th canonical EARS pattern `complex` (while + when) (F-9d168287) (#208)

src/spec/ears.ts implemented 5 of 6 EARS patterns; the 6th — `complex`, a
precondition combined with a trigger ("While A, when B, the system shall X.")
— was missing, and the validator keyed on the first trigger word only, so a
multi-clause `While … when …` requirement either failed validation or was
forced into a single-keyword bucket, silently dropping the trigger clause.

- `EarsPattern` gains `complex`; checkEarsShape validates BOTH clauses (leading
  `while` precondition + a `when` trigger) and names the missing one, preserving
  the precondition→trigger relationship EARS exists to capture.
- Purely additive: the existing five patterns validate exactly as before.
- `complex` mirrored across every enum site in lockstep (types.ts,
  spec/schema.json ac.ears + always_ears, new.ts, MCP server enum,
  spec-conformance message) so HARNESS_INTEGRITY stays green.

Blind-authored tests (tests/spec/ears-complex.test.ts, 14) cover the new
pattern, the missing-clause messages, backward-compat for all 5 prior patterns,
and the schema-mirror.

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

* Knowledge graph — see, ask, and measure how spec, code, tests, and docs connect (#222)

* feat(graph): reverse-edge index (backlinks) — F-ee47fc2b

Derive three reverse maps from the spec's forward edges, memoized per
Spec instance via a WeakMap (no Spec mutation, 0 bytes on disk):
- dependents:        featureId -> direct dependents (inverse depends_on)
- moduleOwners:      module path -> owning feature ids (many-to-many)
- testRefCitations:  test path -> citing feature ids (anchor-stripped)

The backlink seam the graph layer (blast-radius queries, doc linking,
exports) reuses. pruneToFeature only walked depends_on upward; this is
the reverse direction. Tests authored impl-blind (4/4 green).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(graph): blast-radius impact query — F-7794a6bc

buildImpactSlice + clad_get_impact (MCP) + clad impact (CLI): the
backward complement of clad_get_context. Resolve a feature id/slug or a
module path (fanning out through the many-to-many module owners) and walk
the reverse-index dependents to return the blast radius — impacted
features, scenarios at risk, the deduped regression test set, and the
modules in the radius. Bounded by optional depth, deterministic.

Live: `clad impact src/spec/load.ts` → 4 owners, 111 impacted features,
115 regression tests. Logic tested impl-blind (4/4); MCP integration
tested (clad_get_impact). Glossary + verb-list updated for the new surface.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(graph): doc graph — doc↔spec / doc↔doc links + integrity — F-ee5f643e

extractDocReferences scans docs/**/*.md (excluding fixture/benchmark
dirs, stripping code spans) for F-id references and resolved relative
.md links; clad sync materialises spec/_doc-links.yaml (Tier C, the
greppable "which docs explain feature X" index). New DOC_LINK_INTEGRITY
detector (#38): a dead relative .md link is an error; a scoped doc's
unresolved F-id is a warn. Per-file `clad-doc-links: ignore` opt-out for
teaching docs that use illustrative ids (spec-ids-multi-dev, ssot-testing).

The "all documents connected, always current" half of the graph, made
mechanical. Detector green on cladding-self (scoping = zero false-RED).
Logic + detector tested impl-blind (6/6). Detector count 37→38 across
prose claims; A/B reports regenerated for the new info finding.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(graph): knowledge-graph export + hub stats — F-569f4b37

src/graph/: buildGraph materialises ONE typed graph (feature/module/test/
scenario/capability/doc nodes; depends_on/touches/covers/binds/implements/
references/links edges) from spec + reverse-index + doc-links. Renderers:
mermaid + DOT + JSON (stdout) and an Obsidian vault (one note per node with
[[wikilinks]] + Backlinks) — see the graph in best-in-class viewers, no
bespoke UI. `clad graph stats` ranks hubs by degree (what's load-bearing).
`--focus <q> --depth N` exports a legible neighborhood subgraph.

Declared a `graph` foundation layer in architecture.yaml (pure spec reader).
Live: 706 nodes / 1247 edges; top hub src/cli/clad.ts (degree 33). Pure
functions tested impl-blind (6/6). Verb count 19→20; glossary + SVG diagram
counts (→38 detectors) updated.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* chore(graph): link knowledge-graph capability + CHANGELOG + refresh indexes

- spec/capabilities.yaml: new `knowledge-graph` capability (Tier B) grouping
  the four graph features (F-ee47fc2b, F-7794a6bc, F-ee5f643e, F-569f4b37).
- CHANGELOG [Unreleased]: knowledge-graph entry, framed as traceability/
  retrieval (NOT correctness — honest per the A/B record).
- clad sync refresh: spec/index.yaml, spec/_doc-links.yaml, attestation re-stamp.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(graph): bespoke HTML viewer — SSoT 4-tier colors + slug labels — F-02343cd1

The graph, seen our own way (more than the spec asked):
- model: GraphNode.tier (A/B/C/D) classified deterministically — features/
  scenarios=A, capabilities=B, docs by first-line `Cladding · Tier X` banner
  (+ known-filename fallback), modules/tests=code (no tier). Feature labels now
  show the slug (slug ?? title ?? id); full title kept as `detail` for hover.
- render: getTierColor + getTierLegend; tier flows into json, obsidian
  frontmatter, and mermaid (per-tier classDef coloring — bonus).
- viewer-shell + src/graph/viewer/{app.js,styles.css}: toHtmlShell emits ONE
  self-contained, offline, double-clickable .html — a hand-rolled, zero-dep
  canvas force-directed renderer (NOT a vendored 80KB lib: truest to the
  zero-dep ethos, full control, ~no supply-chain surface). Tier colors, slug
  labels, status opacity, degree-sized hubs, a sidebar (search, kind+tier
  filters with counts, tier legend, Calm/Live, theme, labels), hover-
  neighborhood, click-to-pin, zoom/pan, localStorage layout persistence.
- `clad graph export --format html --out f.html` (mandatory --out). Build copies
  viewer assets to dist/viewer/ (schema.json precedent). eslint-ignored.

Live: 710 nodes / 1259 edges → 213KB self-contained .html, byte-deterministic,
zero external requests. Tests authored impl-blind (3/3). gate GREEN.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(graph): live graph view — clad graph serve + clad_get_graph — F-64a5c159

The graph as a LIVE view, not a re-exported snapshot (the user's reframe:
the graph is a pure derivation of the spec, so build the view once and let
it auto-update as development proceeds).

- src/cli/graph-serve.ts: `clad graph serve [--port]` — a stdlib node:http
  server (zero deps). GET / serves the viewer (with an injected SSE reload
  snippet), GET /graph.json recomputes buildGraph on EVERY request (always
  current — no stale-trap), GET /events is a text/event-stream channel.
  node:fs.watch on spec/ + docs/ broadcasts a debounced refresh → open
  browsers auto-reload. Hardened: headersSent-guarded error path +
  closeAllConnections so it shuts down cleanly (and survives EventSource
  disconnects). SSE keep-alive every 30s.
- src/serve/server.ts: clad_get_graph MCP tool — agents read the live
  (optionally focused) graph in one call; never stale.

Live: GET /graph.json reflects the current spec (713 nodes after this
feature landed). Endpoints + broadcast tested impl-blind (2/2); clad_get_graph
tested over MCP. glossary + TOOL_NAMES updated. gate GREEN.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* chore(graph): link viewer + live-serve features to capability + CHANGELOG

- spec/capabilities.yaml: knowledge-graph now also owns the HTML viewer
  (F-02343cd1) and live serve (F-64a5c159).
- CHANGELOG [Unreleased]: the SSoT-tier-colored viewer + the live auto-updating
  serve, framed plainly (see/navigate the structure, not a correctness check).
- clad sync refresh: index.yaml, _doc-links.yaml, attestation re-stamp.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(graph): viewer canvas was unsized (blank graph) — size to viewport + sidebar-aware fit

The canvas had no CSS width/height, so it fell back to the intrinsic 300×150
and sat hidden behind the 264px sidebar — the graph area rendered blank.
Size the canvas to 100vw/100vh and offset fit() by the sidebar width so the
graph centers in the visible area. Verified headless: nodes draw (ctx.arc
fires 31k× across the settle frames on the 713-node self-graph).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(graph): viewer overhaul — radial galaxy + always-on ambient + bloom — F-8234ec3c

Make it extreme, and alive at rest (the user's ask):
- LAYOUT: a degree-weighted radial galaxy. High-degree hubs are pulled toward
  the centre, low-degree leaves toward the rim; charge spreads angularly,
  springs cluster the connected — the graph gathers into one circular galaxy
  with a bright load-bearing core. Verified on the real 717-node graph: 0
  NaN/Inf, hub (deg 35) at dist 412 vs median 987 (central).
- ALIVE AT REST: ambient is now the default — a slow global rotation + node
  breathing + flowing edge particles + hub glow pulse, all O(n)/O(edges) draw
  (rotation is a free transform; physics stays burst-only). "Calm" freezes it.
- LOOK: additive ("lighter") bloom pass so overlapping hubs build a glowing
  core, over a deep-space radial-gradient background. Click-to-focus (persistent
  neighborhood highlight + smooth recenter), smooth view lerp, upright labels.

Plus a headless render test (tests/graph/viewer-render.test.ts): stubs canvas/
document/window, runs the real app.js, and asserts it draws nodes, keeps the
ambient loop alive, and settles to finite positions with the hub central —
guarding the "blank canvas" / NaN-blowup regressions deterministically in-gate.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* chore(graph): link galaxy viewer feature to knowledge-graph capability

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(graph): Obsidian-grade viewer + live conformance heal (killer) — F-04f50847, F-af45042a

Re-architect the viewer to Obsidian quality, and add the killer that only a
spec↔code-connected tool can do.

Obsidian-grade (F-04f50847):
- Continuous low-alpha simulation with an alphaTarget thermostat: dragging a node
  reheats so connected nodes follow elastically (real tension); HOVER pauses the
  sim (motion freezes under the cursor); release decays to rest. Frame-time
  normalized, slow/calm. Removed the global rotation + edge particles + the
  forced radial-by-degree layout (→ organic center+charge+link balance).
- Four force sliders (중심 장력 / 반발력 / 링크 장력 / 링크 거리) live-bound + persisted.
- nodeColor separates all classes: tiers A/B/C/D distinct hues + module(orange)/
  test(green)/doc(pink) distinct (were all gray).

KILLER — live conformance heal (F-af45042a):
- src/stages/graph-health.ts: nodeHealth() runs cladding's drift detectors and
  maps each finding to its graph node (path→module/test/doc, F-id→feature),
  worst-severity per node. (Lives in stages/: graph→stages is forbidden,
  stages→graph is fine.)
- clad graph serve: GET /health.json (live, watch-refreshed); the viewer overlays
  problem nodes (error=red pulse / warn=amber) over the pretty default + an
  in-sync% pill, and heals smoothly on SSE refresh (viewer self-wires events;
  the reload injection is gone). Static export embeds a stamped snapshot.
- Verified live: /health.json flags the exact features whose modules drifted from
  attestation; they heal green once the gate re-attests.

Galaxy viewer (F-8234ec3c) archived — superseded by F-04f50847 (radial layout,
global rotation, particles replaced; bloom + click-focus carried forward).
Headless tests rewritten for the new behavior (hover-pause, drag-reheat, color,
finite) + a health-mapper test. 1534 tests green.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* chore(graph): re-attest after viewer build (heal stale attestation)

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(graph): viewer aesthetics — collision spacing, visible edges, no muddy bloom

The dense overlapping 'confetti' is now a breathing web: a collision pass
hard-separates overlapping nodes (0 overlapping pairs on the 717-node graph,
was a clump), stronger repel + smaller nodes spread it (span ~1350). Edges are
colored-by-source and actually visible (the structure reads); the additive
bloom (which washed the centre muddy/white) is removed; nodes get a thin
bg-colored ring so touching nodes stay crisp; health is a clean pulsing ring
not a blob. Force defaults retuned for an even, Obsidian-like distribution.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(graph): WebGL stellar viewer + working-set context tooling (F-77f7ead0, F-06dfdad6, F-d6b93648)

Knowledge-graph session work on feature/ssot-knowledge-graph:

- webgl-stellar-viewer (F-77f7ead0): real three.js + UnrealBloom 3D galaxy over
  the SSoT graph (semantic hue × degree luminosity) + live drift-health overlay,
  esbuild-bundled offline. Adds a 'skill' node kind (SKILL.md distinct from code).
  Archives the 2D canvas viewer (F-04f50847, superseded); swaps F-af45042a module.
- working-set-assembler (F-06dfdad6): clad_get_working_set — one token-budgeted,
  code-bearing payload (focus + module code + forward needs + backward breaks +
  verify + budget). ~5-8x smaller than reading shard+files; clad_get_context frozen.
- graph-context-wiring (F-d6b93648): PostToolUse auto impact-card after source edits
  + ai_hints/persona nudge so the working set is actually used.

All three: strict gate GREEN + attested; 1603 tests pass; tsc/eslint clean.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(graph): context-efficiency tooling + the depends_on gap fix (F-96250595, F-2be3e3bb, F-15999130, F-16138071)

Four features that make cladding's knowledge graph populated + measurable, on
feature/ssot-knowledge-graph. The arc: 4 A/Bs found the graph tools NULL on agent
correctness — then re-framing to the real goal (search/context efficiency, not
"smarter agents") + fixing the empty-graph root cause turned it positive.

- iterative-impact-slice (F-96250595): buildIterativeImpactSlice — seed→widen→stop
  (coverage/exhaustion/marginal-yield), self-describing depth+stoppedBy+coverage;
  fixes the fixed-depth-1 narrow-miss. Calibrated on real graph data (2 sims).
- infer-depends-on (F-2be3e3bb): clad infer-deps reconstructs feature depends_on
  from the code import graph — the load-bearing graph edge cladding PRODUCED nowhere
  (vapt shipped 0 edges → empty graph → the real cause of the A/B NULLs). vapt 0→698.
- inferable-depends-on detector (F-15999130): INFERABLE_DEPENDS_ON — single info
  finding (never blocks, even strict) surfacing the empty-graph gap so infer-deps
  isn't a latent never-run tool. Detector count 38→39.
- graph-efficiency-measure (F-16138071): clad measure — deterministic per-feature
  search/context efficiency (no agent, no NULL risk). vapt: working-set 4.1x smaller
  context than naive (shard+all modules); dependency radius + regression set resolved
  for free. The goal axis, finally quantified positive.

Honest framing throughout (docs/ab-evaluation/*.md): the graph does NOT make a capable
agent more correct/cheaper (it greps regardless — NULL x4); its value is small context
+ explicit blast radius + known regression set + a queryable dependency graph, for
humans and tools that use it. infer-deps writes nothing (reviewable suggestions only).

All gated: 1624 tests GREEN, tsc/eslint clean, strict gate + attested per feature.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* chore(graph): post-audit hygiene — clarify efficiency-report stats + drop dead StopReason

Read-only adversarial audit of the session found the work correct (4 features
conform, tests honest, claims reproduce, no regressions). Two low-severity hygiene
fixes, zero functional impact:

- docs/ab-evaluation/case-efficiency-measurement.md: the table wrote "3,028 tok vs
  14,442 tok = 4.1x (median)" on one line, which reads as a contradiction (14442/3028
  ≈ 4.8). Both numbers are correct but are DIFFERENT statistics — 4.1x is the median
  of per-feature naive÷slice ratios; 3,028/14,442 are independent medians of slice
  and naive sizes. Split them + added a note so the report can't be misread. (The
  audit verdict marked 4.1x "VERIFIED" and missed this; caught by hand arithmetic.)
- src/optimizer/iterative-slice.ts: removed 'token-budget' from the StopReason union
  — it was declared but never returned (budget overflow stops at max-depth first).
  Synced the test's ALLOWED_STOPS list.

1624 tests GREEN, tsc/eslint clean, strict gate re-attested (F-96250595).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(infer-deps): flag dynamic-import files for honest recall (F-0f2984d0)

The final read-only graph-traversal test on the real doverunner-vapt project
measured inference precision ~100% but recall ~70-75% — the largest remaining gap
being dynamic/runtime imports (importlib.import_module, __import__, getattr-based,
require(<expr>)) that static regex extraction cannot see (e.g. catalog/tool_inventory.py).

Rather than silently under-report those edges, inferDependsOn now collects such
module files into a sorted dynamicImportFiles list so a maintainer knows exactly
which files to review by hand. This is an honest-recall surface, NOT a precision
change — the static edges from the same files are still inferred unchanged (additive).
clad infer-deps prints dynamic_import_files.

Verified: cladding-self flags scripts/build.mjs (require). blind tests 3/3 in a
separate file (existing 6 untouched); 1627 tests GREEN; tsc/eslint clean; strict
gate re-attested.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(graph): color system — hue=kind only, grouped spec/code/test/docs (F-5b188856)

The viewer's node colors had no clear basis: semanticHue let tier win over kind,
so one hue tried to encode two axes and collided — feature and scenario both
rendered blue, tier-C teal ≈ skill turquoise, and the sidebar showed the same
blue labeled both "Spec · sealed" (tier) and "feature" (kind). Tier is derivable
from kind, so encoding both was redundant.

Hue now encodes KIND only. semanticHue ignores tier; tier moves to the sidebar
filter + tooltip. KIND_COL becomes a grouped, simulation-verified palette
(all Y≥125 bloom floor, colorblind 4-group separation, hub-whitening distinctness):
  SPEC = blue family (feature/scenario/capability)
  module = orange (anchor) · test = green (anchor)
  DOCS = pink family (doc/skill) — skill is SKILL.md, a document, moved out of the
  old code-adjacent turquoise.

Sidebar: one color legend grouped into spec/code/test/docs zones; the SSoT tier
section is a swatch-less filter (checkbox + label only, no empty box). Tier labels
simplified to plain words (Spec / Design / Derived / Audit), and code nodes display
as "code" in the viewer (the spec's `modules:` data model is unchanged — display
label only).

Honest residuals (accepted, user-preferred anchors): module orange sits near the
0-45° health-burn arc, disambiguated by the burn's 2.2-3× pulse vs a static node;
orange↔green is the textbook red-green colorblind pair but is conceptually adjacent
(code↔test) with a luminance backstop.

1630 tests GREEN; tsc/eslint clean; strict gate re-attested.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* docs(readme-ko): graph section + live GIF, flat blue-token diagrams, content-accuracy pass

README.ko.html:
- enterprise trust/trace/scale hero; new project-map (graph) section with a live
  galaxy GIF, colour legend, and a clad-graph-serve launch guide
- refreshed numbers (39 detectors / 1630 tests / 196 features, 192 done / v0.7.0)
- unified plain-declarative voice; blue-primary emphasis (green = pass/success only)
- accurate 4-tier SSoT table (intent defined by humans, authored by the LLM per EARS)

docs/img/ko/*.svg (8): unified flat blue-token design system (no shadow/pastel/accent
bars, no version stamps); every claim verified against code — gate triggers 3/9/15 by
cost, authorship human-defines/LLM-writes, segregation-of-duties "aligns with" (not
"maps to") EU AI Act/SOX, Tier-B = project-context.md (not ai_hints), runner examples
= pure executors

docs/img/ko/graph.gif: 880px/8fps live-graph recording (6.9MB)
.gitignore: ignore *.mov (local screen recordings)

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* chore(attest): re-attest after develop merge — refresh verified tree-hashes

The merge moved 66 done features' module trees. A GREEN strict pre-push gate
(type/lint/unit/coverage/deliverable-smoke all pass; STALE_ATTESTATION was the
only outstanding finding) re-verified and re-stamped spec/attestation.yaml.
Drift is now clean.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* docs(readme): sync all 4 READMEs + English diagrams to the finalized ko.html

Brings README.html, README.md, README.ko.md, and docs/img/en/*.svg to parity with
the finalized README.ko.html, and reconciles post-merge counts across every README.

- README.html / README.md — English mirror of ko.html: enterprise hero, the
  project-map (knowledge-graph) section with the live galaxy GIF + colour legend +
  clad-graph-serve launch guide, every section, locked numbers
- README.ko.md — Korean-markdown mirror (plain-declarative voice, docs/img/ko refs)
- docs/img/en/*.svg (8) — English of the finalized ko diagrams (identical geometry/
  style, terse English to fit fixed boxes) + docs/img/en/graph.gif
- counts reconciled everywhere: 40 detectors · 199 features (195 done) · 1664 tests ·
  169 test files · v0.7.0 · 2026-07 (was 196/192, 39, 1630, 2026-06 in places)
- ko.html: card "언제든 추적할 수 있다", Status 196/192 → 199/195, date → 2026-07
- content-accuracy carried to English: SoD "aligns with" (not "maps to") EU AI Act/SOX,
  Tier-B = project-context.md, runner examples = pure executors, gate 3/9/15 by cost,
  knowledge graph = traceability not correctness

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* chore(attest): re-attest after all-READMEs lockstep

5 done features own README/diagram modules touched by the lockstep commit; a GREEN
strict pre-push gate (all other stages pass) re-verified and re-stamped
spec/attestation.yaml. Drift is clean.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(graph): flush stdout before exit — clad graph export truncated piped output at 64 KiB

runGraphExportCommand wrote the rendered graph to stdout then called process.exit(0)
on the next line; on a pipe, stdout.write is async, so exit killed the process before
the ~64 KiB OS pipe buffer drained — `clad graph export --format json | jq` truncated
at exactly 65536 B (this repo's payload is ~285 KiB). File / --out / serve / MCP
(synchronous or HTTP) were unaffected. Exit from the write callback instead; same
prophylactic fix for `graph stats`. Found by the pre-PR empirical verification sweep.

- src/cli/graph.ts: process.stdout.write(x, () => process.exit(0)) at both stdout sites
- tests/cli/graph-export-pipe.test.ts: spawns the CLI through a pipe, asserts the full
  >64 KiB JSON arrives and parses
- count reconciliation from the new test file: tests 1664→1665, test_files 169→170
  across the 4 READMEs + spec.yaml inventory; dist rebuilt; attestation re-stamped

Verified: `clad graph export --format json | jq` → 199 feature nodes (was truncated);
npm test 1665/1665 GREEN; pre-push strict gate green.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* docs(multi-agent): retitle diagram to "agent role separation" (was "persona permission separation")

The multi-agent diagram title used "페르소나 권한 분리" / "Persona Separation of Duties",
but the READMEs call these agents (not personas) and the concept is role/duty separation,
not permissions — and the diagram's own subtitle already says segregation-of-duties.
Aligned the title to that vocabulary:
- ko: "페르소나 권한 분리" → "에이전트 역할 분리"
- en: "Persona Separation of Duties" → "Agent Separation of Duties"
- ko/en multi-agent.svg (title + a11y) + all four README alt texts; attestation re-stamped

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* docs(readme): make README.md + README.ko.md markdown-native (readability)

The two markdown READMEs mirrored the HTML too literally (layout <table>s, an
inline-styled Status block GitHub strips to bare text, centered <p>+<br> prose).
Rewrote them to native markdown — text verbatim, format only:
- layout 3-col tables → stacked sections / bullet trios (the hero hook,
  before/after/record, see/ask/measure)
- Status styled-table → one clean 5-col markdown table
- detector HTML table → markdown table; centered body prose → left-aligned + a blockquote
- one extra blank line before each main (##) section heading
- fixed a pre-existing KO detector-table count (spec↔test 5→6, so the rows sum to 40)
  and added the missing "capability 6개" to the KO Status footnote (EN parity)

HTML siblings (README.html, README.ko.html) and all diagrams left untouched.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(attest): re-stamp at 2026-07-01 — CI stale from last_synced crossing midnight

spec.yaml's inventory.last_synced is a module of F-5b9f9f / F-32b1e0 / F-d6b93648,
so when sync/check advances it to the current date, those features' attested tree-
hash goes stale. The branch was attested on 06-30; CI ran on 07-01 UTC and re-stamped
last_synced to 07-01, so the self-drift gate (`clad check --tier=pre-commit --strict`)
reported STALE_ATTESTATION. Re-attested against today so the committed spec.yaml +
attestation agree with a same-day CI run.

Follow-up (pre-existing, separate): last_synced should not churn attestation — exclude
it from the module hash, or stop writing it from check/build.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* docs(readme): trim prose for concision across all 4 READMEs

The before → verify → record loop was narrated ~5× at different altitudes. Cut the
cross-section redundancy and tightened wordy prose (~-93 lines; larger in rendered prose):
- "how it works with host LLM": dropped the intro + the entire After/Record subsections
  (re-told downstream in Gate / Detector / "done is earned") → one pointer line; kept Before
- removed the duplicate "no commands to memorize" line, the tagline-blurb flourish, the
  inline 9-stage list, and restated cost-split / colour-legend / Authority-column / ecosystem tails

The "기업이~ / To trust AI" tagline + its 3 trust cards are preserved byte-for-byte (core
hook, per maintainer). Numbers, honesty notes, diagrams, commands, and Docs links all intact.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* chore(release): v0.7.0 — version bump (9 sites) + lock + CHANGELOG + rebuild

- version-bump 0.7.0 across the 9 HARNESS_INTEGRITY sites (package.json, marketplace.json,
  both plugin.json, gemini-extension.json, src/cli/clad.ts, src/serve/server.ts,
  tests/cli/clad.test.ts, spec.yaml); package-lock refreshed via npm install
- CHANGELOG: [Unreleased] → [0.7.0] — 2026-07-01 — Knowledge Graph
- dist + plugin mirrors rebuilt; attestation re-stamped
- 1665/1665 tests green; pre-push strict gate green

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Sungju <yuyu04@naver.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

* feat(drift): UNVERIFIED_AC multi-framework test_ref↔testcase matching (closes #203) (#204)

* feat(drift): UNVERIFIED_AC — verify test_refs ran and passed via JUnit XML (F-96700032)

UNTESTED_AC only checks that a done AC's test_refs EXIST on disk — an empty
file, a test.skip, or a failing test all satisfied the gate, so the chain
stopped at "AC → named file", not "AC → observed pass". UNVERIFIED_AC closes
that gap when a JUnit XML report is available.

- New detector UNVERIFIED_AC (src/stages/detectors/unverified-ac.ts) + a pure
  regex JUnit parser (src/stages/junit-report.ts, no XML dependency, mirroring
  the coverage-XML approach). Report path comes from gate.test_report in
  .cladding/config.yaml or a conventional default; failing/errored or
  only-skipped test_refs are an error, absent ones a warn (--strict promotes).
- Graceful by default: no report present → emits nothing, leaving UNTESTED_AC's
  existence check as the baseline, so non-JUnit projects are unaffected.
- Registered in detectors/index.ts (count auto-recomputed 37→38); catalog +
  self-consistency detector-count claims bumped to 38.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* feat(drift): UNVERIFIED_AC multi-framework test_ref↔testcase matching (F-d980359c)

UNVERIFIED_AC's matcher was effectively vitest-only — it keyed every testcase
by `classname` and assumed that was a file path. pytest (`tests.test_foo`),
Java/Kotlin FQCN (`com.example.FooTest`), and `file=`-attribute emitters never
matched, so a passing test read as `absent` (a false positive under --strict)
and a real fail/skip was mis-reported as "did not run".

- parse: index each testcase under every path-shaped key it yields — the
  `file=` attribute, the `classname` as-is, and a dot→slash conversion of a
  dotted (no-separator) classname — sharing one accumulator (no double count).
- lookup: extension-agnostic exact/suffix match (`FooTest` ↔ `FooTest.kt`).
- detector: confident-or-degrade — a report with no path-like keys (e.g. jest
  describe-title classnames) is unmappable, so emit nothing instead of flooding
  false `absent` findings (preserves the low-false-positive contract).

Measured A/B (OLD = classname-only): correct verdicts across a
vitest/pytest/Kotlin/jest matrix 2/8 → 8/8; parse cost on a 10k-case report
+1.6ms (vitest) … +7.8ms (pytest) per gate run — noise vs the ~50s gate.

Blind-authored tests (tests/stages/junit-multiframework.test.ts) cover all 5
ACs. Existing UNVERIFIED_AC tests stay green (backward compatible).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

* feat(ears): add the 6th canonical EARS pattern `complex` (while + when) (F-9d168287) (#208)

src/spec/ears.ts implemented 5 of 6 EARS patterns; the 6th — `complex`, a
precondition combined with a trigger ("While A, when B, the system shall X.")
— was missing, and the validator keyed on the first trigger word only, so a
multi-clause `While … when …` requirement either failed validation or was
forced into a single-keyword bucket, silently dropping the trigger clause.

- `EarsPattern` gains `complex`; checkEarsShape validates BOTH clauses (leading
  `while` precondition + a `when` trigger) and names the missing one, preserving
  the precondition→trigger relationship EARS exists to capture.
- Purely additive: the existing five patterns validate exactly as before.
- `complex` mirrored across every enum site in lockstep (types.ts,
  spec/schema.json ac.ears + always_ears, new.ts, MCP server enum,
  spec-conformance message) so HARNESS_INTEGRITY stays green.

Blind-authored tests (tests/spec/ears-complex.test.ts, 14) cover the new
pattern, the missing-clause messages, backward-compat for all 5 prior patterns,
and the schema-mirror.

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

* Knowledge graph — see, ask, and measure how spec, code, tests, and docs connect (#222)

* feat(graph): reverse-edge index (backlinks) — F-ee47fc2b

Derive three reverse maps from the spec's forward edges, memoized per
Spec instance via a WeakMap (no Spec mutation, 0 bytes on disk):
- dependents:        featureId -> direct dependents (inverse depends_on)
- moduleOwners:      module path -> owning feature ids (many-to-many)
- testRefCitations:  test path -> citing feature ids (anchor-stripped)

The backlink seam the graph layer (blast-radius queries, doc linking,
exports) reuses. pruneToFeature only walked depends_on upward; this is
the reverse direction. Tests authored impl-blind (4/4 green).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(graph): blast-radius impact query — F-7794a6bc

buildImpactSlice + clad_get_impact (MCP) + clad impact (CLI): the
backward complement of clad_get_context. Resolve a feature id/slug or a
module path (fanning out through the many-to-many module owners) and walk
the reverse-index dependents to return the blast radius — impacted
features, scenarios at risk, the deduped regression test set, and the
modules in the radius. Bounded by optional depth, deterministic.

Live: `clad impact src/spec/load.ts` → 4 owners, 111 impacted features,
115 regression tests. Logic tested impl-blind (4/4); MCP integration
tested (clad_get_impact). Glossary + verb-list updated for the new surface.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(graph): doc graph — doc↔spec / doc↔doc links + integrity — F-ee5f643e

extractDocReferences scans docs/**/*.md (excluding fixture/benchmark
dirs, stripping code spans) for F-id references and resolved relative
.md links; clad sync materialises spec/_doc-links.yaml (Tier C, the
greppable "which docs explain feature X" index). New DOC_LINK_INTEGRITY
detector (#38): a dead relative .md link is an error; a scoped doc's
unresolved F-id is a warn. Per-file `clad-doc-links: ignore` opt-out for
teaching docs that use illustrative ids (spec-ids-multi-dev, ssot-testing).

The "all documents connected, always current" half of the graph, made
mechanical. Detector green on cladding-self (scoping = zero false-RED).
Logic + detector tested impl-blind (6/6). Detector count 37→38 across
prose claims; A/B reports regenerated for the new info finding.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(graph): knowledge-graph export + hub stats — F-569f4b37

src/graph/: buildGraph materialises ONE typed graph (feature/module/test/
scenario/capability/doc nodes; depends_on/touches/covers/binds/implements/
references/links edges) from spec + reverse-index + doc-links. Renderers:
mermaid + DOT + JSON (stdout) and an Obsidian vault (one note per node with
[[wikilinks]] + Backlinks) — see the graph in best-in-class viewers, no
bespoke UI. `clad graph stats` ranks hubs by degree (what's load-bearing).
`--focus <q> --depth N` exports a legible neighborhood subgraph.

Declared a `graph` foundation layer in architecture.yaml (pure spec reader).
Live: 706 nodes / 1247 edges; top hub src/cli/clad.ts (degree 33). Pure
functions tested impl-blind (6/6). Verb count 19→20; glossary + SVG diagram
counts (→38 detectors) updated.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* chore(graph): link knowledge-graph capability + CHANGELOG + refresh indexes

- spec/capabilities.yaml: new `knowledge-graph` capability (Tier B) grouping
  the four graph features (F-ee47fc2b, F-7794a6bc, F-ee5f643e, F-569f4b37).
- CHANGELOG [Unreleased]: knowledge-graph entry, framed as traceability/
  retrieval (NOT correctness — honest per the A/B record).
- clad sync refresh: spec/index.yaml, spec/_doc-links.yaml, attestation re-stamp.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(graph): bespoke HTML viewer — SSoT 4-tier colors + slug labels — F-02343cd1

The graph, seen our own way (more than the spec asked):
- model: GraphNode.tier (A/B/C/D) classified deterministically — features/
  scenarios=A, capabilities=B, docs by first-line `Cladding · Tier X` banner
  (+ known-filename fallback), modules/tests=code (no tier). Feature labels now
  show the slug (slug ?? title ?? id); full title kept as `detail` for hover.
- render: getTierColor + getTierLegend; tier flows into json, obsidian
  frontmatter, and mermaid (per-tier classDef coloring — bonus).
- viewer-shell + src/graph/viewer/{app.js,styles.css}: toHtmlShell emits ONE
  self-contained, offline, double-clickable .html — a hand-rolled, zero-dep
  canvas force-directed renderer (NOT a vendored 80KB lib: truest to the
  zero-dep ethos, full control, ~no supply-chain surface). Tier colors, slug
  labels, status opacity, degree-sized hubs, a sidebar (search, kind+tier
  filters with counts, tier legend, Calm/Live, theme, labels), hover-
  neighborhood, click-to-pin, zoom/pan, localStorage layout persistence.
- `clad graph export --format html --out f.html` (mandatory --out). Build copies
  viewer assets to dist/viewer/ (schema.json precedent). eslint-ignored.

Live: 710 nodes / 1259 edges → 213KB self-contained .html, byte-deterministic,
zero external requests. Tests authored impl-blind (3/3). gate GREEN.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(graph): live graph view — clad graph serve + clad_get_graph — F-64a5c159

The graph as a LIVE view, not a re-exported snapshot (the user's reframe:
the graph is a pure derivation of the spec, so build the view once and let
it auto-update as development proceeds).

- src/cli/graph-serve.ts: `clad graph serve [--port]` — a stdlib node:http
  server (zero deps). GET / serves the viewer (with an injected SSE reload
  snippet), GET /graph.json recomputes buildGraph on EVERY request (always
  current — no stale-trap), GET /events is a text/event-stream channel.
  node:fs.watch on spec/ + docs/ broadcasts a debounced refresh → open
  browsers auto-reload. Hardened: headersSent-guarded error path +
  closeAllConnections so it shuts down cleanly (and survives EventSource
  disconnects). SSE keep-alive every 30s.
- src/serve/server.ts: clad_get_graph MCP tool — agents read the live
  (optionally focused) graph in one call; never stale.

Live: GET /graph.json reflects the current spec (713 nodes after this
feature landed). Endpoints + broadcast tested impl-blind (2/2); clad_get_graph
tested over MCP. glossary + TOOL_NAMES updated. gate GREEN.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* chore(graph): link viewer + live-serve features to capability + CHANGELOG

- spec/capabilities.yaml: knowledge-graph now also owns the HTML viewer
  (F-02343cd1) and live serve (F-64a5c159).
- CHANGELOG [Unreleased]: the SSoT-tier-colored viewer + the live auto-updating
  serve, framed plainly (see/navigate the structure, not a correctness check).
- clad sync refresh: index.yaml, _doc-links.yaml, attestation re-stamp.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(graph): viewer canvas was unsized (blank graph) — size to viewport + sidebar-aware fit

The canvas had no CSS width/height, so it fell back to the intrinsic 300×150
and sat hidden behind the 264px sidebar — the graph area rendered blank.
Size the canvas to 100vw/100vh and offset fit() by the sidebar width so the
graph centers in the visible area. Verified headless: nodes draw (ctx.arc
fires 31k× across the settle frames on the 713-node self-graph).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(graph): viewer overhaul — radial galaxy + always-on ambient + bloom — F-8234ec3c

Make it extreme, and alive at rest (the user's ask):
- LAYOUT: a degree-weighted radial galaxy. High-degree hubs are pulled toward
  the centre, low-degree leaves toward the rim; charge spreads angularly,
  springs cluster the connected — the graph gathers into one circular galaxy
  with a bright load-bearing core. Verified on the real 717-node graph: 0
  NaN/Inf, hub (deg 35) at dist 412 vs median 987 (central).
- ALIVE AT REST: ambient is now the default — a slow global rotation + node
  breathing + flowing edge particles + hub glow pulse, all O(n)/O(edges) draw
  (rotation is a free transform; physics stays burst-only). "Calm" freezes it.
- LOOK: additive ("lighter") bloom pass so overlapping hubs build a glowing
  core, over a deep-space radial-gradient background. Click-to-focus (persistent
  neighborhood highlight + smooth recenter), smooth view lerp, upright labels.

Plus a headless render test (tests/graph/viewer-render.test.ts): stubs canvas/
document/window, runs the real app.js, and asserts it draws nodes, keeps the
ambient loop alive, and settles to finite positions with the hub central —
guarding the "blank canvas" / NaN-blowup regressions deterministically in-gate.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* chore(graph): link galaxy viewer feature to knowledge-graph capability

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(graph): Obsidian-grade viewer + live conformance heal (killer) — F-04f50847, F-af45042a

Re-architect the viewer to Obsidian quality, and add the killer that only a
spec↔code-connected tool can do.

Obsidian-grade (F-04f50847):
- Continuous low-alpha simulation with an alphaTarget thermostat: dragging a node
  reheats so connected nodes follow elastically (real tension); HOVER pauses the
  sim (motion freezes under the cursor); release decays to rest. Frame-time
  normalized, slow/calm. Removed the global rotation + edge particles + the
  forced radial-by-degree layout (→ organic center+charge+link balance).
- Four force sliders (중심 장력 / 반발력 / 링크 장력 / 링크 거리) live-bound + persisted.
- nodeColor separates all classes: tiers A/B/C/D distinct hues + module(orange)/
  test(green)/doc(pink) distinct (were all gray).

KILLER — live conformance heal (F-af45042a):
- src/stages/graph-health.ts: nodeHealth() runs cladding's drift detectors and
  maps each finding to its graph node (path→module/test/doc, F-id→feature),
  worst-severity per node. (Lives in stages/: graph→stages is forbidden,
  stages→graph is fine.)
- clad graph serve: GET /health.json (live, watch-refreshed); the viewer overlays
  problem nodes (error=red pulse / warn=amber) over the pretty default + an
  in-sync% pill, and heals smoothly on SSE refresh (viewer self-wires events;
  the reload injection is gone). Static export embeds a stamped snapshot.
- Verified live: /health.json flags the exact features whose modules drifted from
  attestation; they heal green once the gate re-attests.

Galaxy viewer (F-8234ec3c) archived — superseded by F-04f50847 (radial layout,
global rotation, particles replaced; bloom + click-focus carried forward).
Headless tests rewritten for the new behavior (hover-pause, drag-reheat, color,
finite) + a health-mapper test. 1534 tests green.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* chore(graph): re-attest after viewer build (heal stale attestation)

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(graph): viewer aesthetics — collision spacing, visible edges, no muddy bloom

The dense overlapping 'confetti' is now a breathing web: a collision pass
hard-separates overlapping nodes (0 overlapping pairs on the 717-node graph,
was a clump), stronger repel + smaller nodes spread it (span ~1350). Edges are
colored-by-source and actually visible (the structure reads); the additive
bloom (which washed the centre muddy/white) is removed; nodes get a thin
bg-colored ring so touching nodes stay crisp; health is a clean pulsing ring
not a blob. Force defaults retuned for an even, Obsidian-like distribution.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(graph): WebGL stellar viewer + working-set context tooling (F-77f7ead0, F-06dfdad6, F-d6b93648)

Knowledge-graph session work on feature/ssot-knowledge-graph:

- webgl-stellar-viewer (F-77f7ead0): real three.js + UnrealBloom 3D galaxy over
  the SSoT graph (semantic hue × degree luminosity) + live drift-health overlay,
  esbuild-bundled offline. Adds a 'skill' node kind (SKILL.md distinct from code).
  Archives the 2D canvas viewer (F-04f50847, superseded); swaps F-af45042a module.
- working-set-assembler (F-06dfdad6): clad_get_working_set — one token-budgeted,
  code-bearing payload (focus + module code + forward needs + backward breaks +
  verify + budget). ~5-8x smaller than reading shard+files; clad_get_context frozen.
- graph-context-wiring (F-d6b93648): PostToolUse auto impact-card after source edits
  + ai_hints/persona nudge so the working set is actually used.

All three: strict gate GREEN + attested; 1603 tests pass; tsc/eslint clean.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(graph): context-efficiency tooling + the depends_on gap fix (F-96250595, F-2be3e3bb, F-15999130, F-16138071)

Four features that make cladding's knowledge graph populated + measurable, on
feature/ssot-knowledge-graph. The arc: 4 A/Bs found the graph tools NULL on agent
correctness — then re-framing to the real goal (search/context efficiency, not
"smarter agents") + fixing the empty-graph root cause turned it positive.

- iterative-impact-slice (F-96250595): buildIterativeImpactSlice — seed→widen→stop
  (coverage/exhaustion/marginal-yield), self-describing depth+stoppedBy+coverage;
  fixes the fixed-depth-1 narrow-miss. Calibrated on real graph data (2 sims).
- infer-depends-on (F-2be3e3bb): clad infer-deps reconstructs feature depends_on
  from the code import graph — the load-bearing graph edge cladding PRODUCED nowhere
  (vapt shipped 0 edges → empty graph → the real cause of the A/B NULLs). vapt 0→698.
- inferable-depends-on detector (F-15999130): INFERABLE_DEPENDS_ON — single info
  finding (never blocks, even strict) surfacing the empty-graph gap so infer-deps
  isn't a latent never-run tool. Detector count 38→39.
- graph-efficiency-measure (F-16138071): clad measure — deterministic per-feature
  search/context efficiency (no agent, no NULL risk). vapt: working-set 4.1x smaller
  context than naive (shard+all modules); dependency radius + regression set resolved
  for free. The goal axis, finally quantified positive.

Honest framing throughout (docs/ab-evaluation/*.md): the graph does NOT make a capable
agent more correct/cheaper (it greps regardless — NULL x4); its value is small context
+ explicit blast radius + known regression set + a queryable dependency graph, for
humans and tools that use it. infer-deps writes nothing (reviewable suggestions only).

All gated: 1624 tests GREEN, tsc/eslint clean, strict gate + attested per feature.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* chore(graph): post-audit hygiene — clarify efficiency-report stats + drop dead StopReason

Read-only adversarial audit of the session found the work correct (4 features
conform, tests honest, claims reproduce, no regressions). Two low-severity hygiene
fixes, zero functional impact:

- docs/ab-evaluation/case-efficiency-measurement.md: the table wrote "3,028 tok vs
  14,442 tok = 4.1x (median)" on one line, which reads as a contradiction (14442/3028
  ≈ 4.8). Both numbers are correct but are DIFFERENT statistics — 4.1x is the median
  of per-feature naive÷slice ratios; 3,028/14,442 are independent medians of slice
  and naive sizes. Split them + added a note so the report can't be misread. (The
  audit verdict marked 4.1x "VERIFIED" and missed this; caught by hand arithmetic.)
- src/optimizer/iterative-slice.ts: removed 'token-budget' from the StopReason union
  — it was declared but never returned (budget overflow stops at max-depth first).
  Synced the test's ALLOWED_STOPS list.

1624 tests GREEN, tsc/eslint clean, strict gate re-attested (F-96250595).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(infer-deps): flag dynamic-import files for honest recall (F-0f2984d0)

The final read-only graph-traversal test on the real doverunner-vapt project
measured inference precision ~100% but recall ~70-75% — the largest remaining gap
being dynamic/runtime imports (importlib.import_module, __import__, getattr-based,
require(<expr>)) that static regex extraction cannot see (e.g. catalog/tool_inventory.py).

Rather than silently under-report those edges, inferDependsOn now collects such
module files into a sorted dynamicImportFiles list so a maintainer knows exactly
which files to review by hand. This is an honest-recall surface, NOT a precision
change — the static edges from the same files are still inferred unchanged (additive).
clad infer-deps prints dynamic_import_files.

Verified: cladding-self flags scripts/build.mjs (require). blind tests 3/3 in a
separate file (existing 6 untouched); 1627 tests GREEN; tsc/eslint clean; strict
gate re-attested.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(graph): color system — hue=kind only, grouped spec/code/test/docs (F-5b188856)

The viewer's node colors had no clear basis: semanticHue let tier win over kind,
so one hue tried to encode two axes and collided — feature and scenario both
rendered blue, tier-C teal ≈ skill turquoise, and the sidebar showed the same
blue labeled both "Spec · sealed" (tier) and "feature" (kind). Tier is derivable
from kind, so encoding both was redundant.

Hue now encodes KIND only. semanticHue ignores tier; tier moves to the sidebar
filter + tooltip. KIND_COL becomes a grouped, simulation-verified palette
(all Y≥125 bloom floor, colorblind 4-group separation, hub-whitening distinctness):
  SPEC = blue family (feature/scenario/capability)
  module = orange (anchor) · test = green (anchor)
  DOCS = pink family (doc/skill) — skill is SKILL.md, a document, moved out of the
  old code-adjacent turquoise.

Sidebar: one color legend grouped into spec/code/test/docs zones; the SSoT tier
section is a swatch-less filter (checkbox + label only, no empty box). Tier labels
simplified to plain words (Spec / Design / Derived / Audit), and code nodes display
as "code" in the viewer (the spec's `modules:` data model is unchanged — display
label only).

Honest residuals (accepted, user-preferred anchors): module orange sits near the
0-45° health-burn arc, disambiguated by the burn's 2.2-3× pulse vs a static node;
orange↔green is the textbook red-green colorblind pair but is conceptually adjacent
(code↔test) with a luminance backstop.

1630 tests GREEN; tsc/eslint clean; strict gate re-attested.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* docs(readme-ko): graph section + live GIF, flat blue-token diagrams, content-accuracy pass

README.ko.html:
- enterprise trust/trace/scale hero; new project-map (graph) section with a live
  galaxy GIF, colour legend, and a clad-graph-serve launch guide
- refreshed numbers (39 detectors / 1630 tests / 196 features, 192 done / v0.7.0)
- unified plain-declarative voice; blue-primary emphasis (green = pass/success only)
- accurate 4-tier SSoT table (intent defined by humans, authored by the LLM per EARS)

docs/img/ko/*.svg (8): unified flat blue-token design system (no shadow/pastel/accent
bars, no version stamps); every claim verified against code — gate triggers 3/9/15 by
cost, authorship human-defines/LLM-writes, segregation-of-duties "aligns with" (not
"maps to") EU AI Act/SOX, Tier-B = project-context.md (not ai_hints), runner examples
= pure executors

docs/img/ko/graph.gif: 880px/8fps live-graph recording (6.9MB)
.gitignore: ignore *.mov (local screen recordings)

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* chore(attest): re-attest after develop merge — refresh verified tree-hashes

The merge moved 66 done features' module trees. A GREEN strict pre-push gate
(type/lint/unit/coverage/deliverable-smoke all pass; STALE_ATTESTATION was the
only outstanding finding) re-verified and re-stamped spec/attestation.yaml.
Drift is now clean.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* docs(readme): sync all 4 READMEs + English diagrams to the finalized ko.html

Brings README.html, README.md, README.ko.md, and docs/img/en/*.svg to parity with
the finalized README.ko.html, and reconciles post-merge counts across every README.

- README.html / README.md — English mirror of ko.html: enterprise hero, the
  project-map (knowledge-graph) section with the live galaxy GIF + colour legend +
  clad-graph-serve launch guide, every section, locked numbers
- README.ko.md — Korean-markdown mirror (plain-declarative voice, docs/img/ko refs)
- docs/img/en/*.svg (8) — English of the finalized ko diagrams (identical geometry/
  style, terse English to fit fixed boxes) + docs/img/en/graph.gif
- counts reconciled everywhere: 40 detectors · 199 features (195 done) · 1664 tests ·
  169 test files · v0.7.0 · 2026-07 (was 196/192, 39, 1630, 2026-06 in places)
- ko.html: card "언제든 추적할 수 있다", Status 196/192 → 199/195, date → 2026-07
- content-accuracy carried to English: SoD "aligns with" (not "maps to") EU AI Act/SOX,
  Tier-B = project-context.md, runner examples = pure executors, gate 3/9/15 by cost,
  knowledge graph = traceability not correctness

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* chore(attest): re-attest after all-READMEs lockstep

5 done features own README/diagram modules touched by the lockstep commit; a GREEN
strict pre-push gate (all other stages pass) re-verified and re-stamped
spec/attestation.yaml. Drift is clean.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(graph): flush stdout before exit — clad graph export truncated piped output at 64 KiB

runGraphExportCommand wrote the rendered graph to stdout then called process.exit(0)
on the next line; on a pipe, stdout.write is async, so exit killed the process before
the ~64 KiB OS pipe buffer drained — `clad graph export --format json | jq` truncated
at exactly 65536 B (this repo's payload is ~285 KiB). File / --out / serve / MCP
(synchronous or HTTP) were unaffected. Exit from the write callback instead; same
prophylactic fix for `graph stats`. Found by the pre-PR empirical verification sweep.

- src/cli/graph.ts: process.stdout.write(x, () => process.exit(0)) at both stdout sites
- tests/cli/graph-export-pipe.test.ts: spawns the CLI through a pipe, asserts the full
  >64 KiB JSON arrives and parses
- count reconciliation from the new test file: tests 1664→1665, test_files 169→170
  across the 4 READMEs + spec.yaml inventory; dist rebuilt; attestation re-stamped

Verified: `clad graph export --format json | jq` → 199 feature nodes (was truncated);
npm test 1665/1665 GREEN; pre-push strict gate green.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* docs(multi-agent): retitle diagram to "agent role separation" (was "persona permission separation")

The multi-agent diagram title used "페르소나 권한 분리" / "Persona Separation of Duties",
but the READMEs call these agents (not personas) and the concept is role/duty separation,
not permissions — and the diagram's own subtitle already says segregation-of-duties.
Aligned the title to that vocabulary:
- ko: "페르소나 권한 분리" → "에이전트 역할 분리"
- en: "Persona Separation of Duties" → "Agent Separation of Duties"
- ko/en multi-agent.svg (title + a11y) + all four README alt texts; attestation re-stamped

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* docs(readme): make README.md + README.ko.md markdown-native (readability)

The two markdown READMEs mirrored the HTML too literally (layout <table>s, an
inline-styled Status block GitHub strips to bare text, centered <p>+<br> prose).
Rewrote them to native markdown — text verbatim, format only:
- layout 3-col tables → stacked sections / bullet trios (the hero hook,
  before/after/record, see/ask/measure)
- Status styled-table → one clean 5-col markdown table
- detector HTML table → markdown table; centered body prose → left-aligned + a blockquote
- one extra blank line before each main (##) section heading
- fixed a pre-existing KO detector-table count (spec↔test 5→6, so the rows sum to 40)
  and added the missing "capability 6개" to the KO Status footnote (EN parity)

HTML siblings (README.html, README.ko.html) and all diagrams left untouched.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(attest): re-stamp at 2026-07-01 — CI stale from last_synced crossing midnight

spec.yaml's inventory.last_synced is a module of F-5b9f9f / F-32b1e0 / F-d6b93648,
so when sync/check advances it to the current date, those features' attested tree-
hash goes stale. The branch was attested on 06-30; CI ran on 07-01 UTC and re-stamped
last_synced to 07-01, so the self-drift gate (`clad check --tier=pre-commit --strict`)
reported STALE_ATTESTATION. Re-attested against today so the committed spec.yaml +
attestation agree with a same-day CI run.

Follow-up (pre-existing, separate): last_synced should not churn attestation — exclude
it from the module hash, or stop writing it from check/build.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* docs(readme): trim prose for concision across all 4 READMEs

The before → verify → record loop was narrated ~5× at different altitudes. Cut the
cross-section redundancy and tightened wordy prose (~-93 lines; larger in rendered prose):
- "how it works with host LLM": dropped the intro + the entire After/Record subsections
  (re-told downstream in Gate / Detector / "done is earned") → one pointer line; kept Before
- removed the duplicate "no commands to memorize" line, the tagline-blurb flourish, the
  inline 9-stage list, and restated cost-split / colour-legend / Authority-column / ecosystem tails

The "기업이~ / To trust AI" tagline + its 3 trust cards are preserved byte-for-byte (core
hook, per maintainer). Numbers, honesty notes, diagrams, commands, and Docs links all intact.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* Graph capability repairs — the 0.7.0 seams, simulation-verified (#225)

* fix(hygiene): replace raw NUL bytes with \u0000 escapes — sources were binary to git

src/graph/model.ts (edge-dedup key), src/optimizer/infer-depends-on.ts
(edge-map key), src/spec/attestation.ts (sha256 separators) carried literal
0x00 chars inside string/template literals. Git's binary heuristic then hid
those files from every diff/blame/review — the 0.7.0 graph core shipped as
"Bin 0 -> 9363 bytes" with a review-invisible diff. The escape spelling is
byte-identical at runtime (attestation digests unchanged; verified by
simulation before the change).

Adds a self-consistency test banning raw NUL bytes across the source tree
so a file can never silently become review-invisible again.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(graph): legacy F-NNN ids restored to the doc axis via a shared F-id lexer

doc-references.ts matched only 6-8 hex ids, so every legacy sequential id
(F-001…F-083 — 80 live shards) referenced in docs/ produced no doc→feature
edge and no DOC_LINK_INTEGRITY validation, while graph-health.ts already
carried the correct alternation. src/spec/feature-id.ts is now the single
prose-scanning lexer for both sites (fresh RegExp per call — no shared /g
state). Simulated on the live repo before implementing: +10 refs gained,
all 10 resolve to real shards (0 new warns), hash-id extraction
byte-identical, references edges 8→18.

spec/_doc-links.yaml regeneration happens via clad sync at branch close.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(hook): PostToolUse impact card fires on host-style absolute paths

Hosts send tool_input.file_path ABSOLUTE while moduleOwners keys are the
spec's repo-relative posix paths, so buildImpactSlice never resolved and
the impact card (F-d6b93648) never rendered in real usage. Measured on
cladding-self before the fix: 0/361 module paths hit; after relativizing:
358/361 (99.2% — the 3 misses are trailing-slash directory keys unreachable
via the hook). Outside-repo absolute paths degrade to not_found, relative
inputs are untouched (idempotent). Card now also displays the relative path.

Adds the end-to-end wiring test (runHookEvent with an absolute file_path
against a real on-disk spec) that was missing when the bug shipped.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(working-set): module queries seed every co-owner; the budget now caps breaks_if_changed

Two seams in one module, landed together deliberately — simulation showed
the fan-out alone doubles budget-breach paths (17→33/144 on cladding-self,
worst 2.9x the cap), so the clip must ship with it.

Fan-out: a module-path query now passes its original form to the iterative
impact slice, seeding ALL owners (reverse-slice already supported this) —
before, only the alphabetically-first owner's dependents/tests were
reported (src/cli/clad.ts: impacted 0 vs 83). Co-owners are seeds, so they
surface in co_owners, not impacted; feature-id queries are byte-identical.

Clip: breaks_if_changed now participates in the token budget, LAST in the
clipping order (needs → code → breaks; the clip-before-code variant
simulated strictly worse). Deeper dependents drop from the far end first,
then tests outside the depth-1 floor; the direct set is never dropped
(must-edit precedent). Fit checks measure WITH the pending
'breaks: omitted …' marker, closing the +3..10-token overshoot the old
loops carried. In-budget payloads return byte-identical (pure no-op —
existing 6 tests unchanged).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(measure): attribute cap-driven shrink honestly; plumb one reader through both sides

Two validity fixes to clad measure, both simulation-verified before coding:

Attribution: medianShrinkFactor was bounded by the 3000-token default budget
— on cladding-self the '~4x smaller' headline was mostly the CAP's
arithmetic, not graph value (uncapped, the structural slice is ~1.16x of
naive: code + structured metadata). The report now splits
fitsCount/truncatedCount with medianShrinkFit/medianShrinkTruncated and a
medianStructuralRatio, and the CLI headline attributes the reduction to the
budget ('budget enforces 3.9x on 163 capped') instead of selling it as
shrink. What the working set actually sells: a guaranteed token budget +
wired needs/breaks/verify context.

One universe: the injected ModuleReader now reaches buildWorkingSet →
codeExcerpt (same safety gates), so the slice and the naive baseline read
the SAME universe — before, a virtual reader fed only the baseline,
silently inflating the shrink factor ('Pure given (spec, reader)' is now
true). The old test asserted the inflated number; rewritten to assert the
honest split.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(serve): 503 error-as-data on unreadable spec; clean EADDRINUSE; watcher + Host hardening

Body-before-writeHead on every non-SSE route: a mid-write or unparseable
spec throws inside liveGraph(), and committing the 200+application/json
headers first turned that into an HTTP 200 whose body was a prose YAML
error — precisely the state the fs.watch auto-reload refetches into.
Parse failures now answer 503 with a JSON {error} payload (the viewer can
show a retry state); the headersSent guard remains for mid-stream (SSE)
failures. Also covers schema-invalid specs (simulation).

EADDRINUSE: server.on('error') now rejects the boot promise, so
runGraphServeCommand prints one pulse line and exits 1 — before, the
'error' event was unhandled (raw 20-line stack) and the promise never
settled. The listener stays attached so later runtime errors can't crash
the process either. FSWatcher 'error' events degrade to manual refresh
instead of crashing. A foreign Host header is refused (DNS-rebinding
guard; the bind was already 127.0.0.1).

Tests: the old 'watched-file change' test called broadcast() by hand —
renamed to what it tests, and a REAL fs.watch→debounce→SSE chain test
added (writes under spec/, capability-probe skip on platforms without
recursive watch), plus 503, busy-port, and Host-guard cases.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(graph): kind-twin union for focus queries; clad_get_graph summary; health cache + all-twin badges; live viewer fixes

Node identity: the same file materialises as up to three nodes
(module:/test:/doc: — 95 paths on cladding-self), and every focus surface
picked only the FIRST twin, silently dropping the others' edges.
resolveNodeIds returns all twins, subgraph accepts a seed set, and the CLI
--focus + clad_get_graph query union them (resolveNodeId keeps the old
single-id contract). graph-health now badges ALL twins of a finding's path
(first-twin-only left siblings looking healthy) and the viewer's drift
pill counts distinct paths so twins can't double the headline number.

clad_get_graph: the no-query form returned the whole graph pretty-printed
— measured 285KB (~70k tokens) in one MCP result, contradicting the
working-set budget discipline. It now answers a graphStats summary
(counts by kind + top hubs + a hint at the CLI export, 2.1KB); a focused
query still returns the real subgraph. The vacuous clad_get_working_set
test (asserted a hand-maintained constant against itself; the handler was
never invoked) is replaced with a real MCP round-trip: on-disk module code
in must_edit.code, dependent in breaks, budget echo, isError miss.

nodeHealth wraps its detector loop in primeSpecCache (the drift.ts
run-scope pattern): one shard-tree parse instead of ~10 per /health.json
request — measured 611ms → 21ms for the loop, results byte-identical
(incl. mtime-sensitive STALE_TESTS on a drifted fixture).

Viewer: SSE change detection compares the server's exact graph.json text
(baseline = first fetch; never re-serialized client-side) — the old
nodes.length proxy missed edge-only and same-count node changes and
rendered stale. Wires the dead mobile burger button. CLI: a typo'd
--format/--depth now fails loudly instead of silently rendering mermaid.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(infer-deps): extract export…from re-exports + literal dynamic import(); TS fixtures added

The JS/TS branch missed two statically-extractable forms cladding's own
source uses: 'export … from' re-exports (barrel files are dependencies)
and literal dynamic import('…'). Both now produce edges. A NON-literal
import(expr) flags the file in dynamicImportFiles (kept apart from the
shared DYNAMIC_IMPORT regex — 'import (' would false-flag Python's
parenthesized from-import). The whole JS/TS extraction branch shipped with
zero fixtures (all Python) — a TS fixture set now pins import…from,
side-effect import, require(), re-exports, dynamic import, and the
single-segment ambiguity rule.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(render,serve,build): label escaping, corrupt-JSONL tolerance, three.js license notice

Mermaid ids are deduped (safeId collapsed src/a.ts vs src/a_ts into one
node silently); labels flatten quotes/newlines that broke the quoted-label
syntax. DOT escapes backslashes before quotes. Obsidian wikilink aliases
strip the |[[]] metacharacters that corrupt links. clad_get_events
tolerates a corrupt/partial JSONL line (mid-write tail read) as an
{unparseable} entry instead of crashing the tool call. The viewer bundle
keeps three.js's MIT notice (legalComments 'eof' — 'none' stripped it from
a distributed artifact).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* docs: record the viewer decision reversal; correct 0.7.0 claims; cap-attribution note

design.md gains a dated post-ship addendum (§8): the 'no bespoke web UI'
decision was reversed (WebGL viewer + live serve shipped, and why the
premise no longer applied), clad_get_dependents shipped as clad_get_impact,
scope grew (working-set/iterative/infer-deps/measure), and _doc-links.yaml
is a grep/human index, not the export source. The 0.7.0 CHANGELOG described
the archived 2D prototype (force sliders, Live/Calm, 'no third-party
library') — corrected in place with a note, plus an [Unreleased] section
for this branch. Glossary's graph row now names html/serve. README no
longer claims serve 'opens' the browser. case-efficiency-measurement.md
carries the cap-attribution correction (the '4.1×' headline is
budget-enforced, not structural). Deprecation notices say 0.8 (0.7.0
shipped every alias while claiming removal in 0.7).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* spec(graph): shards track the corrected 0.7.1 contracts; sync + build artifacts

Updated ACs/test_refs where behavior contracts changed (clad_get_graph
no-query summary, working-set fan-out + breaks clipping, measure honest
attribution + one-reader universe, hook impact-card absolute-path wiring
now pinned end-to-end), each with a Correction note recording why.
src/spec/feature-id.ts claimed by the doc-graph feature. clad sync
regenerated spec/_doc-links.yaml — the restored legacy F-NNN doc edges
(multi-provider-roadmap, ssot-model) plus design.md's addendum references
now materialize. npm run build refreshed dist + plugin mirrors + the
viewer bundle (SSE text-compare + pill dedupe + burger + three.js MIT
notice at EOF). Persona alias deprecation wording aligned to 0.8.

Gate: clad check --tier=pre-push --strict GREEN (exit 0), full suite
1681/1681.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

* feat(graph): fallback safety contract — a graph answer that cannot know says so (F-c6a32fff) (#226)

The graph layer's miss/exception handling was solid, but three verified
holes let "unknown" read as "safe" — every fix design was simulated
against real repo data before implementation (one proposed fix was
invalidated by the simulation and dropped: promoting clad_get_impact to
the iterative slice would have silently shrunk 41% of responses).

Ledger honesty: every impact slice now carries spec-wide edge counts
{depends_on_edges, test_ref_edges}. On a blank ledger (measured on a
196-feature clone: a feature with 10 real dependents answered
impacted:[] coverage:1.0, byte-identical to a verified leaf) the answer
gains fallback hints — unknown, not safe; fall back to grep / the full
suite. Zero known dependents stops with 'no-known-dependents' +
coverage:null (44% of cladding-self features take this path); the
working-set radius carries the denominator and guards the JS null→0
coercion the simulation caught. The impact card discloses
'· deps unledgered'.

Hook scoping: Stop and PostToolUse now mirror SessionStart's spec.yaml
guard — a non-cladding cwd (or a monorepo subdir; hook cwd is
process.cwd()) used to get falsely BLOCKED by ABSENCE_OF_GOVERNANCE with
.cladding/ state written into the foreign tree (reproduced with the
shipped bundle). Not under cladding → silence, zero writes; a
present-but-broken spec keeps its honest one-time block.

Gate footer: an engine fault fabricated {pass:true} on the one
structural channel hook-less hosts see — now fails closed with
{pass:false, unavailable:true} (pass stays, per the F-570a3f wire
contract). The four graph MCP tools adopt the loadSpecOrError guidance
(was raw ENOENT), clad_get_graph misses gain the discovery hint, and
discovery hints name the baseline fallback. SessionStart renders an
unparseable spec with no resolvable counts as 'counts unavailable'
(conditional — a healthy spec/index.yaml still renders true counts).

Feature cycle: shard F-c6a32fff authored via clad_create_feature, all 6
ACs test_ref-wired, flipped done through the strict pre-push gate.
e2e-verified on the built binary: dense ledger {246,323} no hints /
blank ledger {0,0} both hints / spec-less Stop silent with no .cladding.

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

* Release v0.7.1 — Honest Graph (version bump + CHANGELOG + READMEs)

npm run version-bump -- 0.7.1 (9 sites incl. marketplace catalog) +
package-lock refresh. CHANGELOG [Unreleased] → [0.7.1] — 2026-07-02.
READMEs (md/ko.md/html/ko.html): status v0.7.1, tests 1691/1691,
features 200 (196 done).

Full suite 1691/1691 GREEN; clad check --tier=pre-push --strict GREEN.
Externally validated against npm 0.7.0 on a real 188-feature project:
23 scenarios, 15 measured improvements, 0 regressions.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

---------

Co-authored-by: Sungju <yuyu04@naver.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>