@nurix/etna 0.19.88 → 0.19.89

package/data/harness-index.json

@@ -1,5 +1,5 @@
 {
-    "fingerprint": "417d61b8de2d",
+    "fingerprint": "f3094aa08430",
     "harnesses": {
         "code": {
             "plugin": "code",

package/data/harnesses/discover/CLAUDE.md

@@ -86,7 +86,7 @@ Cheap, read-only. Scope the target, grade its access, detect which dimensions it
 
 Dispatch one collector per available dimension; collect exhaustively before context is lost. Every collector writes `research/<target>/dimensions/<dim>/_summary.md` (with the **provenance frontmatter** — method · confidence · completeness · gaps) + `raw/`, obeying `.claude/rules/ingestion.md`.
 
-- **Run independents in parallel** — dispatch `docs`, `packages`, `api`, `website`, `community`, `deployed-client-bundle`, `infra-backend-fingerprint`, `distribution-artifacts` together (`subagent_type='<dimension>'`, one message, multiple Agent calls). Seed each collector with **warm entry points (URLs / repos / hosts), not claimed facts** — any fact **or inference** carried from Discovery (a price, a tier name, a version — **or a derived hypothesis like "localStorage-key X ⇒ backend is Y"**) is tagged `verify:may-be-stale` so the collector **re-derives it instead of anchoring** on a possibly-stale (or possibly-wrong) value. A Discovery _inference_ is never seeded as a plan fact: label it a hypothesis in `00-recon-plan.md` and tag it `verify:` in the seed, so a collector that inherits it reports it hedged rather than echoing it as observed (a real run seeded an `apolloAnonId ⇒ GraphQL backend` guess that two captures then repeated before the session refuted it). **A sibling collector's _summary_ is a claim, not a source** — when seeding one collector with something another collector reported, tag it `verify:` too, unless the main session has actually read the underlying `raw/` artifact. A returned summary is a compressed, unverified assertion; relaying it into a fresh brief launders it into apparent provenance, and the receiving sub-agent (which cannot see the conversation) has no way to tell a seeded claim from an established fact. (Origin: Emergent — the orchestrator relayed a "documented limitation: the connector cannot deploy" line from one collector's summary into another's brief; it was **not in the corpus**, and only the receiving agent's own verification caught it.) Only one dimension may drive the browser at a time (see `ingestion` §7 rule 9): the parallel collectors use WebFetch/curl, leaving Chrome to `session`.
+- **Run independents in parallel** — dispatch `docs`, `packages`, `api`, `website`, `community`, `deployed-client-bundle`, `infra-backend-fingerprint`, `distribution-artifacts` together (`subagent_type='<dimension>'`, one message, multiple Agent calls). Seed each collector with **warm entry points (URLs / repos / hosts), not claimed facts** — any fact **or inference** carried from Discovery (a price, a tier name, a version — **or a derived hypothesis like "localStorage-key X ⇒ backend is Y"**) is tagged `verify:may-be-stale` so the collector **re-derives it instead of anchoring** on a possibly-stale (or possibly-wrong) value. A Discovery _inference_ is never seeded as a plan fact: label it a hypothesis in `00-recon-plan.md` and tag it `verify:` in the seed, so a collector that inherits it reports it hedged rather than echoing it as observed (a real run seeded an `apolloAnonId ⇒ GraphQL backend` guess that two captures then repeated before the session refuted it). **A sibling collector's _summary_ is a claim, not a source** — when seeding one collector with something another collector reported, tag it `verify:` too, unless the main session has actually read the underlying `raw/` artifact. A returned summary is a compressed, unverified assertion; relaying it into a fresh brief launders it into apparent provenance, and the receiving sub-agent (which cannot see the conversation) has no way to tell a seeded claim from an established fact. (Origin: Emergent — the orchestrator relayed a "documented limitation: the connector cannot deploy" line from one collector's summary into another's brief; it was **not in the corpus**, and only the receiving agent's own verification caught it.) **Every QUANTITATIVE claim in a brief carries its lane.** The rule above covers a *sibling's summary*; it does not cover a number the **orchestrator itself** picked up in passing — from a search result, a page it skimmed, or its own recollection — and then re-emitted as framing. A figure the main session cannot name a `raw/` anchor for is either seeded **`verify:`** or **omitted**; it is never stated flat. A number in a brief reads as established even when nothing established it, and the receiving agent cannot tell the difference. (Origin: indus-sarvam — the orchestrator seeded a "₹98.68 crore IndiaAI funding" figure into the positioning brief as context; the receiving agent traced it to a **third-party user's GitHub issue body**, not to any first-party source, and had to de-launder it. The vendor states only "compute provided under the IndiaAI mission", with no amount.) Only one dimension may drive the browser at a time (see `ingestion` §7 rule 9): the parallel collectors use WebFetch/curl, leaving Chrome to `session`.
 - **Codebase fan-out** — dispatch `codebase`; it clones + maps + returns a per-package work-list; then the main session fans out one reader per substantial package.
 - **Gated dimensions** — `session` and `wire-capture` (rung ≥2) need explicit user authorization first (see `ingestion` §7). `session` follows `.claude/rules/ingestion.md` §9. When `session` is authorized, **dispatch the non-browser collectors in the background (`run_in_background`) and drive the gated browser session in the main session concurrently** — they don't contend (Chrome belongs to the main session; the collectors use WebFetch/curl), so the whole of Ingestion runs in parallel.
 - **Shared seam** — **seed the two shared files `dimensions/_shared/{api-path-catalog.md, feature-flags.md}` (empty, with a header) once, before the collector fan-out**; bundle/session/wire/ distribution then each _append_ a `source:`-tagged section to the same physical file (never their own copy). One owner dimension per static asset; a second mine of the same asset is one source, not corroboration (`ingestion` §6).

package/data/harnesses/discover/harness.json

@@ -1,12 +1,12 @@
 {
   "harness": "discover",
   "name": "discover",
-  "version": "0.1.15",
+  "version": "0.1.16",
   "files": [
     ".prettierignore",
     ".prettierrc.json",
     ".vscode/settings.json"
   ],
   "seeds": [],
-  "digest": "b55e0ea4a713"
+  "digest": "0bf7a6a6f479"
 }

package/data/harnesses/discover/rules/cartography-coverage.md

@@ -111,7 +111,10 @@ experiential / IA coverage axis (`self-correction.md` rubric, axis 2, max 20) is
   unlocated capabilities and hiding them would inflate the rate. But they are **dispositioned**, so they
   never draw the −2 un-dispositioned penalty, and the scorecard carries `features_not_yet_reachable` so a
   reader can tell a *policy/authorization* boundary from a *thoroughness* failure.
-- `flow_coverage` = `100 × (primary journeys diagrammed in ux-flows.md ÷ journeys identified)`.
+- `flow_coverage` = `100 × (primary journeys diagrammed in ux-flows.md ÷ journeys identified)` — **read both
+  numbers from `ux-flows.md`'s `journeys_diagrammed` / `journeys_identified` frontmatter**
+  (`cartography-flows.md`), never re-derive the denominator here. A reconstructed denominator lets the
+  scorer pick the number that flatters the score, and makes the metric non-reproducible across runs.
 - `screenshot_coverage` = `100 × (primary surfaces with a screenshot ÷ all primary surfaces)`.
 - `ia_nav_complete` = every D0 nav destination has a surface card.
 

package/data/harnesses/discover/rules/cartography-flows.md

@@ -60,6 +60,15 @@ job id, a streamed result). Stop at the terminal write or external response.
 
 ````markdown
 # <Target> — UX Flows
+---
+journeys_identified: <N>   # REQUIRED — every primary journey the corpus identifies, whether or not it
+                           # was diagrammed. This is the DENOMINATOR of feature-coverage's
+                           # `flow_coverage`; without it the metric is reconstructed by whoever scores
+                           # it, and a run that identifies few journeys scores the same as one that
+                           # identifies many and diagrams them all.
+journeys_diagrammed: <N>   # the count of sequence diagrams below
+undiagrammed: [<name>, …]  # the named difference — never an empty list with N_identified > N_diagrammed
+---
 ## TL;DR                  (how the product gets work done: sync-REST vs async-job, where realtime lives,
                            the approval/gate pattern, what's brokered server-side)
 ## Flow index             (table: flow · category · trigger · participants · complexity)

package/data/harnesses/discover/rules/ingestion.md

@@ -78,11 +78,11 @@ Keep `status` and `confidence` honest — they drive the entire evaluation weigh
 
 ## 3. Method & confidence vocabulary (drives evaluation weighting)
 
-`method` is one of: `clone-and-map`, `source-map-reassembly`, `registry-metadata`, `ats-api`, `openapi-verbatim`, `graphql-introspection`, `crawl-clip`, `docs-reconstructed`, `bundle-string-mine`, `dns-ct-fingerprint`, `wire-tap-browser`, `wire-har`, `wire-mitmproxy`, `binary-extract` (asar/apk/crx), `session-playbook`, `sdk-harness`, `inferred`.
+`method` is one of: `clone-and-map`, `source-map-reassembly`, `registry-metadata`, `ats-api`, `first-party-api-json`, `openapi-verbatim`, `graphql-introspection`, `crawl-clip`, `docs-reconstructed`, `bundle-string-mine`, `dns-ct-fingerprint`, `wire-tap-browser`, `wire-har`, `wire-mitmproxy`, `binary-extract` (asar/apk/crx), `session-playbook`, `sdk-harness`, `inferred`.
 
 **Confidence is a function of method** (evaluation relies on this mapping):
 
-- **high** — verbatim machine artifacts: `openapi-verbatim`, `graphql-introspection`, `registry-metadata`, `ats-api`, `clone-and-map`, `binary-extract`.
+- **high** — verbatim machine artifacts: `openapi-verbatim`, `graphql-introspection`, `registry-metadata`, `ats-api`, `first-party-api-json`, `clone-and-map`, `binary-extract`.
 
 > **`ats-api` — a first-party job-board / ATS JSON API is NOT a crawl.** A response from an official
 > applicant-tracking endpoint (`boards-api.greenhouse.io/v1/boards/<slug>/jobs?content=true`,
@@ -94,6 +94,20 @@ Keep `status` and `confidence` honest — they drive the entire evaluation weigh
 > (Origin: Emergent — a 433 KB verbatim Greenhouse JSON body carrying a byte-identical declared tech stack
 > across two postings was labelled `crawl-clip` with `confidence: high`, a mismatch against this very
 > table. The confidence was right and the vocabulary had no word for the artifact.)
+
+> **`first-party-api-json` — the general case of the rule above.** `ats-api` fixed the *hiring* instance
+> of a wider gap: **any verbatim JSON response from a first-party or official platform API is a machine
+> artifact, not a crawl**, and maps to **high**. Use it for a GitHub org/issues/releases API response
+> (`api.github.com/orgs/<org>/repos`, `/search/issues?q=org:<org>`), a status-page API
+> (`status.<vendor>.com/api/v2/summary.json`), a changelog/release JSON feed, a public app-store listing
+> API, or any comparable official endpoint. **The distinction is the artifact, not the topic:** the
+> *rendered page* of the same information is `crawl-clip` (medium); the *API body* is
+> `first-party-api-json` (high). Prefer the more specific term where one exists — `registry-metadata` for
+> npm/PyPI/crates, `ats-api` for job boards, `openapi-verbatim` for a served spec — and reach for this one
+> only when none fits. (Origin: indus-sarvam — the `community` lane mined the GitHub API and, finding no
+> correct term, graded itself `registry-metadata`. The evidence class and the `high` confidence were both
+> right; the label named the wrong registry. A vocabulary with no word for an artifact does not stop a
+> collector using it — it just makes the provenance record lie.)
 - **medium** — `crawl-clip`, `docs-reconstructed`, `bundle-string-mine`, `dns-ct-fingerprint`, `wire-har`.
 - **low** — `inferred`, `session-playbook` Pass-1-only, `wire-tap-browser` shapes-only, any wire-drift observation without a spec.
 
@@ -284,7 +298,7 @@ the elements needing redaction change as you navigate.
 
 `deployed-client-bundle` (static unauth SPA), `session` (authenticated browser runtime), and `wire-capture` (non-browser client) can all surface the same **API path catalog** and **feature flags**. To keep them consumable instead of triplicated **N times** (a real run forked `api-path-catalog.md` up to 7×, and "two copies agree" then mis-reads as corroboration):
 
-- **One physical file, one canonical location, append-only.** The seam files live at `dimensions/_shared/api-path-catalog.md` and `dimensions/_shared/feature-flags.md` — **Ingestion seeds the two empty files (with a header) once, before the collector fan-out.** Every contributing dimension **appends a `source:`-tagged section** (`source: bundle | session | wire | distribution`) and **never creates its own copy**; each collector's `_summary.md` Artifacts list **links** the shared file rather than duplicating it. **Shared-seam allowlist:** these two filenames are the _only_ shared/single-physical artifacts — every other `raw/*` is dimension-private.
+- **One physical file, one canonical location, append-only.** The seam files live at `dimensions/_shared/api-path-catalog.md` and `dimensions/_shared/feature-flags.md` — **Ingestion seeds the two empty files (with a header) once, before the collector fan-out.** Every contributing dimension **appends a `source:`-tagged section** (`source: bundle | session | wire | distribution | api`) and **never creates its own copy**; each collector's `_summary.md` Artifacts list **links** the shared file rather than duplicating it. **Shared-seam allowlist:** these two filenames are the _only_ shared/single-physical artifacts — every other `raw/*` is dimension-private.
 - Evaluation owns the **API-path diff** (an _adaptive N-way_ diff over whichever runtime sources were actually run — see evaluation template 3) in `data-model-api-surface.md`. A path that appears in `bundle` but never in `session`/`wire` is a dormant/unshipped route — a finding, not noise.
 - **The cloned source tree is a shared working copy too (a sequencing race, not just the seam files).** On a `source:full` target, `codebase` clones into `source/<target>/`, and other dimensions (`docs`/`packages`) may read in-repo files from it. A clone-dependent collector dispatched **concurrently** with the clone can read a **mid-checkout tree** (a `.git`-only state) and silently fall back to a weaker source. So either (a) treat the `codebase` clone as a **sequencing barrier** — let it finish before dispatching any collector that reads `source/<target>/` — or (b) require each clone-dependent collector to **fetch its own copy** (GitHub API / its own shallow clone) and never assume the shared clone is ready. (Origin: Twenty — `docs` raced the `codebase` clone, saw a `.git`-only tree, and fell back to the live docs site; `packages` correctly fetched its own manifests and was unaffected.)
 
@@ -294,7 +308,9 @@ the elements needing redaction change as you navigate.
 >
 > **Paired inputs make the diff feedable (capture cue).** The split is only emittable if BOTH lanes were captured: when a target has **a published API AND an authed session**, the **pre-nav in-page fetch/XHR tap** (`session` — the app-own paths it actually calls) and the **unauth bundle/route-manifest** (`deployed-client-bundle` — the published paths) are **paired required inputs**. Install the tap _before_ navigation (else on-load app calls are missed) and capture the manifest unauth; a session that only screenshots, or a bundle mine with no session, leaves the diff un-feedable. (Origin: Akeneo — 54 session-observed app-own paths vs a 124-path unauth `/api/rest/v1` manifest, near-disjoint.)
 >
-> **Host-level dimensions cite, they don't append (seam allowlist).** The shared `api-path-catalog.md` `## source:` allowlist is **bundle / session / wire / distribution** only — the path-contributing runtime lanes. A host-level dimension (`infra-backend-fingerprint`) that wants to cross-reference the catalog does so **from its own `raw/`** (cite hosts/paths by reference), and does **not** append its own `## source:` section to the shared file. (Origin: Akeneo — infra appended a host-level note to the path catalog; benign but blurs the allowlist.)
+> **Host-level dimensions cite, they don't append (seam allowlist).** The shared `api-path-catalog.md` `## source:` allowlist is **bundle / session / wire / distribution** (the path-contributing *runtime* lanes) **plus `api`** (the *published* lane — see below). A host-level dimension (`infra-backend-fingerprint`) that wants to cross-reference the catalog does so **from its own `raw/`** (cite hosts/paths by reference), and does **not** append its own `## source:` section to the shared file. (Origin: Akeneo — infra appended a host-level note to the path catalog; benign but blurs the allowlist.)
+>
+> **`api` IS on the allowlist — but it is the PUBLISHED lane, and it never corroborates a runtime path.** The published developer/partner spec belongs in the same physical file as the runtime lanes because the **published-vs-app-own diff** (evaluation template 3) is the whole point of the seam, and it is far easier to compute when both sides sit side by side under one header. Two hard constraints ride with the inclusion: (a) an `## source: api` section carries the **spec-declared** surface only — never a path the collector merely inferred; and (b) **`api` is excluded from the promotion rule for app-own paths** — a path appearing in `## source: api` and in `## source: session` is *the published surface and the runtime surface agreeing*, which is a **finding about drift (or its absence)**, not two independent dimensions corroborating one claim. (Origin: indus-sarvam — the `api` lane appended a 70-operation published catalog in violation of the then-allowlist; deleting it would have made the corpus strictly worse, because that section is exactly what made the published-vs-app-own diff feedable. The rule was wrong, not the collector.)
 
 ---
 
@@ -559,7 +575,27 @@ A single linear sweep wastes context on things you can read for free and wastes
 
 **Pass 2 — fresh-run validation (with explicit cost/state-change confirmation).** Only after Pass 1, and only with explicit user OK stating the exact action + cheapest mode + estimated cost. Catches what Pass 1 can't: the submit/write request body, live state-machine transitions, the completion event, server-side resolution of shorthand, ID/slug-generation, the real iteration endpoint, and the full realtime event catalog. **Legitimate to skip** if the user opts out — leave write-side shapes as open questions; never fabricate them.
 
-> **Record the write-side decision as a machine flag, not re-derived prose.** A Pass-1-only run leaves the _entire_ write/generate/launch surface unobserved — feature _existence/gating_ is solid, but feature _execution behavior_ rests on marketing/docs and must never be promoted to fact (this caveat otherwise gets re-hedged across 3–4 rollups every cost-gated run). So the session `_summary.md` frontmatter carries `write_side_observed: <true|false>` and `pass2: <not-applicable | offered-pending | declined-by-user | deferred | done>` (this flag is the load-bearing part — it lets Evaluation render the write-side gap _once_). When it would help a later run resume, optionally add `captures/pass2-followup.md` listing the **exact cheapest-mode write actions + est. cost** that would close the gap. Discovery seeds the `pass2:` decision in `00-recon-plan.md` for any `auth:have-login` target.
+> **Record the write-side decision as a machine flag, not re-derived prose.** A Pass-1-only run leaves the _entire_ write/generate/launch surface unobserved — feature _existence/gating_ is solid, but feature _execution behavior_ rests on marketing/docs and must never be promoted to fact (this caveat otherwise gets re-hedged across 3–4 rollups every cost-gated run). So the session `_summary.md` frontmatter carries `write_side_observed: <true|false>` and `pass2: <not-applicable | offered-pending | declined-by-user | deferred | authorized-declined | done>` (this flag is the load-bearing part — it lets Evaluation render the write-side gap _once_).
+
+> **An AUTHORIZED Pass-2 that is never exercised is a DEFECT, not a default.** When the user has granted
+> Pass-2 (explicitly, or by saying the run may spend credits / test it), the session lane must do one of
+> exactly two things before Evaluation:
+>
+> 1. **Exercise the cheapest write action** that closes the gap — and there is almost always a cheap one:
+>    a simulated/dry run, a draft save, a version bump, a single smallest-unit generation. Price it, name
+>    it, run it, capture the request+response envelope.
+> 2. **Record `pass2: authorized-declined` with the reason** (the user wrapped the run early; the cheapest
+>    action still mutated shared state; no non-destructive write exists), plus the exact actions and costs
+>    that would close it.
+>
+> `pass2: deferred` is **not** a valid terminal state for an authorized run — it is the state that lets an
+> authorization quietly expire. Mode 5 treats an authorized-but-unexercised Pass-2 as a **medium Ingestion
+> defect** and charges it, because a run that holds permission to observe the write side and ships without
+> it has left its single largest blind spot open by inattention rather than by policy. (Origin:
+> indus-sarvam — the user granted credit spend mid-run; the orchestrator acknowledged it, planned around
+> it, asked which action to take, and shipped `write_side_observed: false` when no reply arrived. The
+> whole write surface, the deploy transition and the live-call realtime plane went unobserved with 100
+> credits unspent.) When it would help a later run resume, optionally add `captures/pass2-followup.md` listing the **exact cheapest-mode write actions + est. cost** that would close the gap. Discovery seeds the `pass2:` decision in `00-recon-plan.md` for any `auth:have-login` target.
 
 **Pass 3 — synthesis.** Walk the captures, pull each finding into the dimension's `_summary.md`, note the cleverest design choices, defaults imposed, performance characteristics, and open questions.
 

package/data/harnesses/discover/rules/tradecraft.md

@@ -128,6 +128,40 @@ pbpaste > capture.json      # macOS
 xclip -o  > capture.json    # Linux
 ```
 
+> **⚠ VERIFY THE PASTE — `copied=true` is NOT proof the OS clipboard changed.** `execCommand('copy')`
+> reports success from *inside the page* and can still leave the system clipboard holding a **previously
+> copied payload** — most reliably when Chrome is not the OS-focused window (the agent's terminal is).
+> A naive `pbpaste > raw/x.json` then writes **the wrong body** into `raw/` under a confident `_meta`,
+> which is the worst artifact class there is: a stale file that scores silently.
+>
+> **Always assert the length matches what the tap reported**, before trusting the file:
+>
+> ```bash
+> EXPECT=14866                      # the len= the __CP call returned
+> GOT=$(pbpaste | wc -m | tr -d ' ')   # -m = CHARACTERS, never -c (bytes). See below.
+> [ "$GOT" -ge $((EXPECT-100)) ] && [ "$GOT" -le $((EXPECT+100)) ] \
+>   && pbpaste > raw/capture.json && echo "ok len=$GOT" \
+>   || echo "STALE CLIPBOARD: expected ~$EXPECT, got $GOT — do NOT write"
+> ```
+>
+> **Compare CHARACTERS (`wc -m`), not BYTES (`wc -c`).** JavaScript's `String.length` counts UTF-16 code
+> units; `wc -c` counts UTF-8 bytes. Any non-ASCII payload makes the two diverge badly — Devanagari,
+> Bengali, Tamil, CJK and emoji are 3–4 bytes per character, so a perfectly fresh 16,378-character capture
+> measures 19,740 bytes and a byte-based guard **rejects a good artifact**. Allow ±100 for the trailing
+> newline and minor encoding slack. (Origin: indus-sarvam — this guard's own first use produced a false
+> positive on a Hindi-language agent transcript; the clipboard was fine and the check was wrong. On a
+> multilingual target the byte form fails *more often than it succeeds*.)
+>
+> Cheap corroboration: also compare a head token (`pbpaste | head -c 40`) against the head the tap
+> returned. **Escalation ladder when it is stale:** (1) click inside the page and retry — it is flaky, not
+> permanently broken; (2) retry once more with the textarea left mounted a beat longer; (3) fall back to
+> paging via `javascript_tool` slices (budget ~1 KB of *usable* output per slice, not the 3 KB the slice
+> size suggests); (4) if paging is uneconomic, write a **decoded** artifact with `verbatim: false`,
+> `_meta.why_not_verbatim` naming the failure, and `_meta.refetch` carrying the exact command — an honest
+> decoded artifact beats a confident wrong one. (Origin: indus-sarvam — a 14,866-byte flag payload
+> reported `copied=true` three times while `pbpaste` kept returning a 195,216-byte body copied minutes
+> earlier. Caught only because the length was checked.)
+
 **Scrub before copying**, not after: build the object with `__SCRUB` first, then `window.__CP(pack)`. The
 clipboard is local, so this stays inside the no-exfiltration rule (ingestion §7 rule 7).
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@nurix/etna",
-  "version": "0.19.88",
+  "version": "0.19.89",
   "description": "Bootstrap a project with a Nurix harness — copies the harness's CLAUDE.md, skills, agents, rules, and hooks directly into the project from a bundled payload (no marketplace fetch, no network).",
   "type": "module",
   "bin": {
@@ -42,5 +42,5 @@
     "agent",
     "plugin"
   ],
-  "etnaFingerprint": "417d61b8de2d"
+  "etnaFingerprint": "f3094aa08430"
 }