feat(audit): attribution on --base added rows

.agents/skills/harden-pr/SKILL.md

@@ -2,7 +2,7 @@
 name: harden-pr
 description: >-
   Bring a branch to pristine, maximum production readiness without changing PR intent —
-  spawn parallel reviewer subagents, fix in-bounds findings, loop autonomously until
+  spawn parallel Task subagents (never inline review), fix in-bounds findings, loop autonomously until
   clean or pass cap, then report once. Use after a tracer-bullet commit (lite), before PR
   is done (full), or on "harden", "harden-pr", "pristine", "review until clean",
   "production-ready pass". Invoking this skill authorizes one harden commit at cycle end.
@@ -33,12 +33,12 @@ Sister skills: [`audit-pr-architecture`](../audit-pr-architecture/SKILL.md) (ext
 
 Otherwise: resolve anchor → run all passes → fix → verify → next pass → finish → report.
 
-| Phase           | Behavior                                                                                                                              |
-| --------------- | ------------------------------------------------------------------------------------------------------------------------------------- |
-| **During loop** | Autonomous. Spawn reviewers in parallel, merge findings, fix in-bounds, re-run checks, advance pass counter.                          |
-| **After loop**  | Single concise report: mode, passes run, production-bar status (met / gaps), fixes made, checks status, deferred nits (if any).       |
-| **Commit**      | If there are uncommitted fixes: one `harden: …` commit **without asking** — skill invocation authorizes it. If no fixes: skip commit. |
-| **Babysit**     | Full mode only. One line at end of report: "For GitHub/CI, run `/babysit`." Do not ask.                                               |
+| Phase           | Behavior                                                                                                                                        |
+| --------------- | ----------------------------------------------------------------------------------------------------------------------------------------------- |
+| **During loop** | Autonomous. `Task`-batch reviewers in parallel (§ Spawning subagents), merge JSON findings, fix in-bounds, re-run checks, advance pass counter. |
+| **After loop**  | Single concise report: mode, passes run, production-bar status (met / gaps), fixes made, checks status, deferred nits (if any).                 |
+| **Commit**      | If there are uncommitted fixes: one `harden: …` commit **without asking** — skill invocation authorizes it. If no fixes: skip commit.           |
+| **Babysit**     | Full mode only. One line at end of report: "For GitHub/CI, run `/babysit`." Do not ask.                                                         |
 
 ## Modes
 
@@ -84,11 +84,116 @@ Reviewers treat the anchor as contract. Findings that would violate it → **rep
 
 **Do not defer complements:** agent-surface parity (rule/skill/MCP), glossary/architecture/golden-queries contracts, script/golden tests for acceptance criteria, and cross-links named in the plan ship in the **same PR** — not "optional v2" or post-merge unless the plan **Out of scope** section explicitly excludes them.
 
+## Spawning subagents (non-negotiable)
+
+The parent agent **MUST NOT** perform reviewer duties inline. Every pass **starts** with a parallel `Task` batch; grep/read/diff by the parent is setup only, **not** a substitute for reviewers.
+
+| Rule            | Requirement                                                                                                          |
+| --------------- | -------------------------------------------------------------------------------------------------------------------- |
+| **Tool**        | Cursor **`Task`** tool only (`subagent_type`: `generalPurpose` for most reviewers; `explore` for Structure lite)     |
+| **Batching**    | One parent message → **all** applicable reviewers for that pass **in parallel** (single turn, multiple `Task` calls) |
+| **Readonly**    | `readonly: true` on every reviewer `Task` — reviewers report; parent fixes                                           |
+| **Pass credit** | A pass counts only after parent merges subagent JSON and acts on actionable findings                                 |
+
+**Anti-pattern (invalid harden):** parent reads the diff, runs tests, fixes nits, and reports — without spawning the roster below.
+
+### Finding schema (every reviewer returns this)
+
+Each reviewer returns **only** a JSON array (no prose wrapper). Parent parses arrays from all reviewers, then merges.
+
+```json
+{
+  "finding": "One-sentence claim about a gap vs production bar",
+  "severity": "blocker | major | minor | nit | info",
+  "file": "repo-relative/path or \"multiple\"",
+  "fixable_in_bounds": true
+}
+```
+
+**Severity → action**
+
+| Severity                   | Parent action                                                       |
+| -------------------------- | ------------------------------------------------------------------- |
+| `blocker` / `major`        | Fix in pass 1; must fix or defer with plan Out of scope before done |
+| `minor` / `nit`            | Fix when in touched files; pass 2+ if pass 1 was crowded            |
+| `info`                     | Log only unless zero-cost fix in diff                               |
+| `fixable_in_bounds: false` | Final report deferred list — do not apply                           |
+
+**Merge + dedupe (parent, after each batch)**
+
+1. Concatenate all reviewer arrays.
+2. Drop `info` unless it blocks ship shape.
+3. Dedupe: same `file` + same root cause → keep highest severity, merge `finding` text.
+4. Sort actionable: `blocker` → `major` → `minor` → `nit`.
+5. If merged list is empty → pass succeeds; skip fix phase.
+
+**Example merged queue (pass 1)**
+
+```json
+[
+  {
+    "finding": "CLI --help documents summary counts but not per-row attribution on --base JSON rows.",
+    "severity": "major",
+    "file": "src/cli/cmd-audit.ts",
+    "fixable_in_bounds": true
+  },
+  {
+    "finding": "Skill shard leaks requiredColumns when describing attribution.",
+    "severity": "major",
+    "file": "templates/agent-content/skill/10-recipes-context.md",
+    "fixable_in_bounds": true
+  },
+  {
+    "finding": "No e2e test for attribution: inherited on deprecated delta.",
+    "severity": "nit",
+    "file": "src/application/audit-worktree.test.ts",
+    "fixable_in_bounds": true
+  }
+]
+```
+
+### Reviewer prompt template (copy per `Task`)
+
+Fill `{ROLE}`, `{REPO}`, `{INTENT_ANCHOR}`, `{SCOPE}`, `{EXTRA}`; set `subagent_type` and `readonly: true`.
+
+```text
+You are the **{ROLE}** reviewer for `/harden-pr` on `{REPO}`.
+
+**Intent anchor (contract — do not suggest changes that violate):**
+{INTENT_ANCHOR}
+
+**Scope:** {SCOPE}
+(lite: slice diff files; full: `git diff --name-status origin/main...HEAD`)
+
+**Production bar:** See harden-pr skill § Production bar — optimize for {ROLE} rows.
+
+**Task:** {EXTRA}
+
+**Return ONLY** a JSON array of findings:
+[{ "finding": "...", "severity": "blocker|major|minor|nit|info", "file": "...", "fixable_in_bounds": true|false }]
+If clean: []
+
+Readonly — do not edit files.
+```
+
+**`{EXTRA}` by role**
+
+| Role               | `{EXTRA}`                                                                                                          |
+| ------------------ | ------------------------------------------------------------------------------------------------------------------ |
+| Correctness        | Read changed source + tests; run affected `bun test <files>`; bugs, edge cases, missing coverage                   |
+| Ship-readiness     | Grep inbound refs to deleted plan files; verify plan retired + lifted; changeset consumer-clean; cross-ref anchors |
+| Structure (lite)   | Read `docs/architecture.md` § Layering; check diff imports for boundary violations; optional codemap queries       |
+| Consumer surface   | Read `consumer-surfaces` rule; parity across CLI help, MCP description, agent-content, README, changeset           |
+| Structure (full)   | Run `audit-pr-architecture` skill read-only; report only fixable-in-bounds items                                   |
+| Schema / migration | `SCHEMA_VERSION`, migration paths, column contract drift                                                           |
+| Security           | auth, secrets, env, user-input paths in diff                                                                       |
+| Performance        | hot paths, benchmarks, worker pools in diff                                                                        |
+
 ## Reviewer roster
 
-Spawn applicable reviewers **in parallel** via subagents in **one batch per pass**. Each returns `{ finding, severity, file, fixable_in_bounds }`.
+Spawn applicable reviewers **in parallel** via **`Task`** in **one batch per pass**. Each subagent returns the finding schema above.
 
-### Core (always)
+### Core (always — every pass)
 
 1. **Correctness** — gaps vs production bar; bugs, edge cases, missing tests in changed paths
 2. **Ship-readiness** — gaps vs production bar; docs, changesets, consumer-surface leaks, error messages; **grep inbound refs → delete shipped plan file → lift to `golden-queries.md` / `architecture.md` / `roadmap.md`**; run [`verify-after-each-step`](../../rules/verify-after-each-step.md) checks on touched files
@@ -114,8 +219,8 @@ Execute **without pausing for user input** until exit condition:
 resolve intent anchor
 pass = 1
 loop:
-  spawn reviewers (parallel, one batch)
-  merge + dedupe findings
+  Task-batch all applicable reviewers (parallel, readonly)
+  parent: merge + dedupe JSON findings (§ Finding schema)
   if none actionable → goto done
   fix in-bounds (pass 1: all; passes 2+: blockers first, then in-scope nits)
   run project checks on touched files

.changeset/audit-delta-attribution.md

@@ -0,0 +1,5 @@
+---
+"@stainless-code/codemap": patch
+---
+
+On `audit --base <ref>` (CLI / MCP / HTTP), each `added` row carries `attribution: introduced | inherited` (branch-new vs pre-existing at merge base). `--summary` adds `added_introduced` / `added_inherited` per delta.

README.md

@@ -122,7 +122,7 @@ codemap audit --json --summary --baseline base                  # counts-only 
 codemap audit --files-baseline base-files                       # explicit per-delta — runs only the slots provided
 codemap audit --baseline base --files-baseline hotfix-files     # mixed — auto-resolve deps + deprecated; override files
 codemap audit --baseline base --no-index                        # skip the auto-incremental-index prelude (frozen-DB CI)
-codemap audit --base origin/main --json                         # ad-hoc — archive+reindex against any committish; no --save-baseline needed
+codemap audit --base origin/main --json                         # ad-hoc — added[].attribution; --summary adds added_introduced/inherited
 codemap audit --base origin/main --format sarif                 # emit SARIF 2.1.0 directly (Code Scanning); also: --ci alias
 codemap audit --base origin/main --ci                           # CI shortcut: --format sarif + non-zero exit on additions
 codemap audit --base v1.0.0 --files-baseline pre-release-files  # mix --base with per-delta override