AC-859: Clean the stale instrument LLM prompt/TODO path (#1223)

* chore(ac-859): delete the unused opt-out prompt and stale Layer 8 TODOs

The per-file opt-out tip prompt (spec site 2) was exported from prompts.ts
and re-exported through the barrel but never wired: only site 1 (rationale)
and site 3 (session summary) reach enhance(). Delete FILE_OPT_OUT_TIP_PROMPT
+ FileOptOutTipContext, their barrel re-exports, and their test block.

The pr-body-renderer TODO(A2-I Layer 8) markers said 'replace with
enhancer.enhance(...)', but both sites already call enhance({ defaultNarrative,
... }) with the default template as the fallback, so the hookup is done.
Remove the stale TODOs and the header line claiming TODO markers remain; note
that site 2 is spec-described but unimplemented. npm run lint passes.

* chore(ac-859): finish removing stale Layer 8 commentary (PR #1223 review)

Two comments still described the pre-Layer-8 state: the pr-body-renderer
section header called these 'the three LLM enhancement sites' (only two are
implemented), and the instrument barrel said Layer 8 'lands next; its hooks
are wired as no-ops ... with TODO markers.' Layer 8 has landed and the TODOs
were removed in this PR. Update both to match. Comment-only.

Author: jayscambler

ts/src/control-plane/instrument/index.ts

@@ -1,9 +1,9 @@
 /**
  * Public barrel for A2-I `autoctx instrument` tool infrastructure.
  *
- * Layers 1 + 2 + 3 + 4 + 5 + 6 + 7 — contract + scanner + safety + registry +
- * planner + pipeline + cli. (Layer 8 — LLM enhancer — lands next; its hooks
- * are wired as no-ops in pipeline/pr-body-renderer.ts with TODO markers.)
+ * Layers 1 through 8: contract, scanner, safety, registry, planner, pipeline,
+ * cli, and the LLM enhancer. The enhancer wires `enhance()` at the two
+ * implemented narrative sites in pipeline/pr-body-renderer.ts.
  *
  * Name-collision resolution:
  *   - `parseDirectives` is exported from BOTH `safety/` (canonical Buffer form)

ts/src/control-plane/instrument/llm/index.ts

@@ -3,10 +3,8 @@
  */
 export {
   RATIONALE_PROMPT,
-  FILE_OPT_OUT_TIP_PROMPT,
   SESSION_SUMMARY_PROMPT,
   type RationaleContext,
-  type FileOptOutTipContext,
   type SessionSummaryContext,
 } from "./prompts.js";
 

ts/src/control-plane/instrument/llm/prompts.ts

@@ -7,10 +7,12 @@
  * `--prompt-template-dir` override (spec §2 deferred items) but NOT shipped
  * in A2-I.
  *
- * All three sites described in spec §10.1:
+ * Prompts ship for the two enhancement sites A2-I implements (spec §10.1):
  *   1. Per-call-site rationale (italic line under each before/after snippet)
- *   2. Per-file opt-out tip (hint box when a file looks unusual)
  *   3. Session summary (top-of-`pr-body.md` paragraph)
+ *
+ * Site 2 (per-file opt-out tip) is described in the spec but not implemented
+ * in A2-I; its prompt is intentionally absent rather than shipped unused.
  */
 
 export interface RationaleContext {
@@ -21,13 +23,6 @@ export interface RationaleContext {
   readonly afterSnippet: string;
 }
 
-export interface FileOptOutTipContext {
-  readonly filePath: string;
-  readonly language: string;
-  /** Heuristic signals detected about this file (e.g. "looks-like-test-file"). */
-  readonly heuristicSignals: readonly string[];
-}
-
 export interface SessionSummaryContext {
   readonly filesAffected: number;
   readonly callSitesWrapped: number;
@@ -66,42 +61,14 @@ export function RATIONALE_PROMPT(ctx: RationaleContext): string {
   ].join("\n");
 }
 
-/**
- * Per-file opt-out tip prompt (spec §10.1 site 2).
- *
- * Suggests an opt-out path when a file looks unusual (test file not in
- * excludes, synthetic-traffic generator, etc). Output is a single short
- * hint (one or two sentences) that will surface in a dedicated hint box
- * in `pr-body.md`.
- */
-export function FILE_OPT_OUT_TIP_PROMPT(ctx: FileOptOutTipContext): string {
-  const signals = ctx.heuristicSignals.length
-    ? ctx.heuristicSignals.join(", ")
-    : "none";
-  return [
-    "You are a helpful coding assistant reviewing an instrumentation plan.",
-    `File: ${ctx.filePath}`,
-    `Language: ${ctx.language}`,
-    `Heuristic signals: ${signals}`,
-    "",
-    "This file looks unusual for instrumentation. Write a single short hint",
-    "(one or two sentences) suggesting how to opt out if that wasn't intended.",
-    "Mention both path-level (`.gitignore` or `--exclude`) and file-level",
-    "(`# autocontext: off-file`) approaches. Keep it actionable and terse.",
-    "Output only the hint prose — no preamble, no markdown headings.",
-  ].join("\n");
-}
-
 /**
  * Session summary prompt (spec §10.1 site 3).
  *
  * Asks for a one-paragraph overview for the top of `pr-body.md`. Highlights
  * anything notable (e.g., "two files were skipped due to secret literals").
  */
 export function SESSION_SUMMARY_PROMPT(ctx: SessionSummaryContext): string {
-  const plugins = ctx.registeredPluginIds.length
-    ? ctx.registeredPluginIds.join(", ")
-    : "(none)";
+  const plugins = ctx.registeredPluginIds.length ? ctx.registeredPluginIds.join(", ") : "(none)";
   return [
     "You are writing a one-paragraph summary of an autocontext instrumentation session.",
     `Files affected: ${ctx.filesAffected}`,

ts/src/control-plane/instrument/pipeline/pr-body-renderer.ts

@@ -5,13 +5,12 @@
  * the output is machine-parseable by downstream CI-review tooling.
  *
  * LLM enhancement discipline (spec §10):
- *   - Three narrative sites total: per-call-site rationale, per-file opt-out
- *     tips, session summary.
- *   - A2-I wires a static default for EACH site. Layer 8 replaces the
- *     `defaultRationale`/`defaultSummary` calls with LLM-enhanced variants
- *     when `enhancer.enhance(default, ctx)` returns non-null; on any failure
- *     the default is used silently.
- *   - This file contains TODO markers at each site for the Layer 8 hookup.
+ *   - Two narrative sites are implemented: per-call-site rationale and session
+ *     summary. (Spec site 2, the per-file opt-out tip, is not implemented.)
+ *   - Each site computes a static default (`defaultRationale`/`defaultSummary`)
+ *     and passes it to `enhance({ defaultNarrative, ... })`, which returns the
+ *     LLM-enhanced variant when enabled and non-empty, or the default on any
+ *     failure or when disabled.
  *
  * Byte-determinism (spec §9.4):
  *   - `pr-body.md` is NOT byte-deterministic when LLM enhancement is enabled.
@@ -281,14 +280,13 @@ export async function renderPrBody(inputs: PrBodyInputs): Promise<string> {
 }
 
 // ---------------------------------------------------------------------------
-// Default narrative templates (the three LLM enhancement sites)
+// Default narrative templates (the two LLM enhancement sites)
 // ---------------------------------------------------------------------------
 
 /**
  * Default single-paragraph session summary. Grouping by SDK keeps readers
- * oriented when multiple plugins ran in one invocation.
- *
- * TODO(A2-I Layer 8): replace with `enhancer.enhance(defaultSummary(ctx), ctx)`.
+ * oriented when multiple plugins ran in one invocation. Used as the
+ * `defaultNarrative` fallback for the session-summary enhancement site.
  */
 function defaultSummary(inputs: PrBodyInputs): string {
   const sdkCounts = new Map<string, number>();
@@ -307,9 +305,8 @@ function defaultSummary(inputs: PrBodyInputs): string {
 
 /**
  * Default per-call-site rationale. Spec §10.4 says narrative explains what
- * the change does + why; the default is terse-but-accurate.
- *
- * TODO(A2-I Layer 8): replace with `enhancer.enhance(defaultRationale(ctx), ctx)`.
+ * the change does + why; the default is terse-but-accurate. Used as the
+ * `defaultNarrative` fallback for the per-call-site rationale enhancement site.
  */
 function defaultRationale(file: PerFileDetailedEdits, edit: EditDescriptor): string {
   const sdk = file.sdkBreakdown[0]?.sdkName ?? "LLM client";

ts/tests/control-plane/instrument/llm/prompts.test.ts

@@ -1,7 +1,6 @@
 import { describe, test, expect } from "vitest";
 import {
   RATIONALE_PROMPT,
-  FILE_OPT_OUT_TIP_PROMPT,
   SESSION_SUMMARY_PROMPT,
 } from "../../../../src/control-plane/instrument/llm/prompts.js";
 
@@ -34,29 +33,6 @@ describe("RATIONALE_PROMPT", () => {
   });
 });
 
-describe("FILE_OPT_OUT_TIP_PROMPT", () => {
-  test("includes heuristic signals and mentions both opt-out mechanisms", () => {
-    const out = FILE_OPT_OUT_TIP_PROMPT({
-      filePath: "tests/test_llm.py",
-      language: "python",
-      heuristicSignals: ["looks-like-test-file"],
-    });
-    expect(out).toContain("tests/test_llm.py");
-    expect(out).toContain("looks-like-test-file");
-    expect(out).toMatch(/\.gitignore|--exclude/);
-    expect(out).toMatch(/autocontext: off/);
-  });
-
-  test("handles empty heuristic signals list", () => {
-    const out = FILE_OPT_OUT_TIP_PROMPT({
-      filePath: "x.py",
-      language: "python",
-      heuristicSignals: [],
-    });
-    expect(out).toContain("none");
-  });
-});
-
 describe("SESSION_SUMMARY_PROMPT", () => {
   test("includes counts and plugin list", () => {
     const out = SESSION_SUMMARY_PROMPT({