@mrciphersmith/keryx 0.2.105 → 0.2.107

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mrciphersmith/keryx",
-  "version": "0.2.105",
+  "version": "0.2.107",
   "description": "Version-controlled project context for AI coding agents: code graph, architecture wiki, project memory, relevant tests, quality signals, and task flows.",
   "private": false,
   "publishConfig": {

package/src/gdskills/bundled/skills/review/review-layout/SKILL.md

@@ -227,7 +227,7 @@ Report done only once all of these hold:
 - The measurement was taken at the worst real case — the longest translation in
   the catalog, the narrowest supported host, the state where the sibling is
   absent — not at the convenient one.
-- Every finding names `file:line` on the component that renders the box.
+- Every finding names the `file` of the component that renders the box and `quote`s the code it is about; `line` is derived at ingest from that quote, never counted.
 - Every `blocker` and `major` carries `class_scope` with `sites` and an
   `enumeration_method` naming the search that derived the set of elements sharing
   the sizing shape.

package/src/gdskills/bundled/skills/review/review-orchestrator/SKILL.md

@@ -1350,10 +1350,10 @@ Each reviewer must return a `REVIEW_RESULT` object matching `.metaproject/skills
 
 Before consolidation, validate every reviewer result:
 - Required status: `DONE`, `DONE_WITH_CONCERNS`, `NEEDS_CONTEXT`, or `BLOCKED`.
-- Required finding fields: id, severity, file, line (nullable only for repo-wide findings), problem, impact, suggested_fix, evidence, confidence, reviewer.
+- Required finding fields: id, severity, file, `quote`, problem, impact, suggested_fix, evidence, confidence, reviewer. **`quote`, not `line`:** copy the code the finding is about verbatim out of `file` and do not count lines — `keryx review ingest` locates the quote at the round's head and derives `line` from where it matched, so a number reported here is overwritten. A model is poor at counting lines and good at repeating text it has just read; this asks only for the second. Quote enough to be unambiguous: a snippet appearing twice in the file is recorded `unlocatable` rather than anchored to the first hit. A finding about the round rather than a site carries neither.
 - Every blocker must include evidence and a concrete suggested fix.
 - Findings without evidence are downgraded to `info` or returned to the reviewer for clarification.
-- Duplicate findings are merged by `dedupe_key` or by `(file, line, problem)`.
+- Duplicate findings are merged by `dedupe_key` or by `(file, quote, problem)`.
 - `NEEDS_CONTEXT` triggers one targeted context refill. If still unresolved, keep it as an explicit open question, not as a blocker.
 - If a reviewer exceeds `max_findings`, keep blockers/majors first and summarize lower severity findings.
 

package/src/gdskills/bundled/skills/review/review-orchestrator/reviewer-finding.schema.json

@@ -74,12 +74,20 @@
             ]
           },
           "line": {
+            "description": "Do not count lines. Quote the code in `quote` instead and leave this null: `keryx review ingest` finds the quote in `file` and writes the line the match started on, overwriting anything reported here. A model is poor at counting lines and good at repeating text it has just read, and this asks it only for the second. Kept in the contract because a finding with nothing to quote — one about the round rather than a site — still has nowhere else to point.",
             "type": [
               "integer",
               "null"
             ],
             "minimum": 1
           },
+          "quote": {
+            "description": "The code this finding is about, copied verbatim out of `file`. A few lines is enough — it is matched against the file at the round's head, exactly first and then with whitespace collapsed. A quote that appears in more than one place is recorded as `unlocatable` rather than anchored to the first hit, so quote enough to be unambiguous.",
+            "type": [
+              "string",
+              "null"
+            ]
+          },
           "problem": {
             "type": "string"
           },

package/src/gdskills/contracts/review-finding.schema.json

@@ -100,7 +100,40 @@
       "description": "Which repository `file` and `line` are in. Null, the default, means the repository under review. A cross-repo claim names the producer here and pins the ref it was read at in `evidence`; without this field a finding about another service's behaviour could not say where its evidence lived, so the rule that such a claim is `info` unless it has a `file:line` at a pinned ref could not be checked by anything but the reviewer's word. Declared in BOTH this schema and skills/review/review-orchestrator/reviewer-finding.schema.json: this one is additionalProperties:false, so a finding carrying `repo` is rejected here if only the reviewer-side schema knows about it."
     },
     "file": { "type": ["string", "null"] },
-    "line": { "type": ["integer", "null"], "minimum": 1 },
+    "line": {
+      "title": "Where the finding points — DERIVED when the finding quotes code",
+      "description": "Reported by the reviewer only when there is nothing to quote. When `quote` is present this field is OUTPUT: `review ingest` finds the quote in `file` at the round's head and writes the line the match started on, overwriting whatever the reviewer said and recording both in `locator`. The reason is the failure it replaces — nothing compared a reported line to anything, so a number that had drifted travelled into findings.json, into the report, and into the disposition recorded against it. On a fix round the file has moved under the finding by construction, which is when the reported number is least trustworthy and most acted upon.",
+      "type": ["integer", "null"],
+      "minimum": 1
+    },
+    "quote": {
+      "title": "The code this finding is about, copied verbatim out of `file`",
+      "description": "What replaces asking the model for a line number: a model is poor at counting lines and good at repeating text it has just read. The quote is matched against `file` at the round's head — exactly first, then with whitespace collapsed, then not at all. A quote that matches in more than one place is `unlocatable` rather than anchored to the first hit, because choosing among them would be a guess wearing a line number. Absent for a finding that is about the round rather than a site.",
+      "type": ["string", "null"]
+    },
+    "locator": {
+      "title": "How `line` was arrived at",
+      "description": "Present exactly when the finding carries a quote. `derived` means the quote was found and `line` is where; `unlocatable` means it was not, and then `line` is null rather than a number nobody checked. `reported_line` keeps what the reviewer said, so a round can measure how far the claims drifted from the code — which is the only way to know whether deriving them repaired anything or is insurance.",
+      "type": "object",
+      "additionalProperties": false,
+      "required": ["state"],
+      "properties": {
+        "state": { "type": "string", "enum": ["derived", "unlocatable"] },
+        "method": { "type": "string", "enum": ["exact", "whitespace-normalised"] },
+        "reason": { "type": ["string", "null"] },
+        "reported_line": { "type": ["integer", "null"], "minimum": 1 }
+      },
+      "allOf": [
+        {
+          "if": { "properties": { "state": { "const": "derived" } }, "required": ["state"] },
+          "then": { "required": ["method"] }
+        },
+        {
+          "if": { "properties": { "state": { "const": "unlocatable" } }, "required": ["state"] },
+          "then": { "required": ["reason"] }
+        }
+      ]
+    },
     "symbol": { "type": ["string", "null"] },
     "problem": { "type": "string" },
     "impact": { "type": "string" },