fix(communication): resolve prompt from split helper

TODO.md

@@ -2,6 +2,11 @@
 
 ## Active tickets
 
+- [ ] [`ticket-033`](project/ticket-033/README.md) — restore communication LLM
+  prompt resolution after its helper moved one directory deeper. Current
+  state: `IN_PROGRESS / VALIDATION`; focused tests pass and both targeted
+  failures are removed, with no test or prompt-content changes.
+
 - [ ] [`ticket-019`](project/ticket-019/README.md) — publish the dependency-free
   Python SDK as the root PyPI distribution `todo2code` through `goal -a`, with
   one root `pyproject.toml` and SDK-only artifacts. Current state:

project/TICKETS.md

@@ -27,4 +27,5 @@ of `project/README.md`, which remains a generated technical-analysis artifact.
 | **ticket-019** | [`README.md`](./ticket-019/README.md) | [`preprompt.md`](./ticket-019/preprompt.md) | - |  [`ai-codex.md`](./ticket-019/ai-codex.md) |  [`ai-codex-logs.txt`](./ticket-019/ai-codex-logs.txt) | [`changelog.md`](./ticket-019/changelog.md) |
 | **ticket-020** | [`README.md`](./ticket-020/README.md) | [`preprompt.md`](./ticket-020/preprompt.md) | - |  [`ai-codex.md`](./ticket-020/ai-codex.md) |  [`ai-codex-logs.txt`](./ticket-020/ai-codex-logs.txt) | [`changelog.md`](./ticket-020/changelog.md) |
 | **ticket-022** | [`README.md`](./ticket-022/README.md) | [`preprompt.md`](./ticket-022/preprompt.md) | - |  [`ai-codex.md`](./ticket-022/ai-codex.md) |  [`ai-codex-logs.txt`](./ticket-022/ai-codex-logs.txt) | [`changelog.md`](./ticket-022/changelog.md) |
+| **ticket-033** | [`README.md`](./ticket-033/README.md) | [`preprompt.md`](./ticket-033/preprompt.md) | - |  [`ai-codex.md`](./ticket-033/ai-codex.md) |  [`ai-codex-logs.txt`](./ticket-033/ai-codex-logs.txt) | [`changelog.md`](./ticket-033/changelog.md) |
 <!-- AUTO:TICKET_INDEX:END -->

project/ticket-033/README.md

@@ -0,0 +1,63 @@
+# Ticket 033: Restore communication prompt resolution after module split
+
+- **ID**: ticket-033
+- **Owner**: unresolved:human
+- **Status**: IN_PROGRESS
+- **Workflow state**: VALIDATION
+- **Created**: 2026-08-04
+
+## Goal and scope
+
+Restore runtime resolution of `communication-to-intent.system.md` after the
+communication LLM implementation moved one directory deeper. The copied path
+logic still climbs three levels, which resolves from
+`dist/src/communication/llm` to `dist/prompts`; repository prompts live at the
+root `prompts` directory, requiring four levels from the compiled helper.
+
+The repair changes only the relative prompt root in the helper. It will not
+copy prompts into `dist`, change provider fallback behavior, weaken
+`require-llm`, edit tests or contact a live provider. One production file is
+allowed and expected implementation time is under 30 minutes.
+
+## Planned changed paths
+
+- `src/communication/llm/implementation-helpers.ts`: correct the prompt root
+  for the module's new filesystem depth.
+- `project/ticket-033/**`, `TODO.md`, `project/TICKETS.md`: intent and evidence.
+
+## Acceptance criteria
+
+- [x] AC-01: Scope is approved by a human owner.
+- [x] AC-02: The communication prompt resolves from both compiled test/runtime
+      execution without relying on stale artifacts.
+- [x] AC-03: Both audited communication enrichment tests pass using their
+      mocked provider and no live request.
+- [x] AC-04: `require-llm` and correction validation behavior remain strict.
+- [x] AC-05: Clean build, focused/full tests, diff scope and governance evidence
+      are recorded without changing tests or prompt contents.
+
+## Participants
+
+- Human participant: unresolved; no user-* file was created by this script.
+- Agent participant: [ai-codex.md](ai-codex.md)
+
+## Approval boundary
+
+- Current state: `IN_PROGRESS / VALIDATION` after implementing and testing the
+  bounded repair.
+- Chat approval authorizes this interactive implementation only; trusted merge
+  evidence must remain independent and bound to the final SHA.
+
+## Validation evidence
+
+- `npm run build`: PASS.
+- Focused communication LLM suite: 3 passed, 0 failed; mocked provider only.
+- Full fresh suite on isolated base: 338 tests, 328 passed, 9 failed and one
+  skipped. Both prompt failures are removed. The remaining failures correspond
+  exactly to independent tickets 030, 031 and 032.
+- `git diff --check`: PASS.
+- `make governance`: only four inherited ticket-018/ticket-019 diagnostics;
+  no ticket-033 scope or workstream finding.
+- Ephemeral integration of implementation commits `9c25942`, `bbd236f`,
+  `c5db5fb` and `f96c0a8` over base `caf6551`: clean build PASS; full suite
+  338 tests, 337 passed, 0 failed and one environment-dependent JDK skip.

project/ticket-033/ai-codex-logs.txt

@@ -0,0 +1,14 @@
+2026-08-04 planning
+- Fresh base `caf6551`: two communication LLM tests fail because the split
+  helper resolves the prompt under nonexistent `dist/prompts`.
+- No production source edited; state remains `WAIT_FOR_APPROVAL`.
+- Human response `kontynuuj` authorized ticket-033; transitioned to
+  `IN_PROGRESS / EDIT`.
+- `npm run build`: PASS.
+- `node --test dist/test/communication-llm.test.js`: PASS (3/3), mocked provider.
+- Fresh full suite: 338 total, 328 pass, 9 fail, 1 skip; both ticket-033 prompt
+  failures removed, remaining failures belong to tickets 030/031/032.
+- `git diff --check`: PASS.
+- `make governance`: four inherited ticket-018/ticket-019 errors, no ticket-033 finding.
+- Combined implementation-only integration on `caf6551`: build PASS; full suite
+  PASS (338 total, 337 pass, 0 fail, 1 JDK skip).

project/ticket-033/ai-codex.md

@@ -0,0 +1,45 @@
+---
+participant-id: agent:codex
+participant: codex
+role: agent
+ticket: ticket-033
+---
+# Participant: codex (AI agent)
+
+## Understanding
+
+The old implementation lived at `src/communication/llm.ts`; its compiled
+module needed three parent traversals to reach repository-root `prompts`.
+`loadCommunicationPrompt` now lives in
+`src/communication/llm/implementation-helpers.ts`, one directory deeper, but
+retained the old traversal. The error therefore names nonexistent
+`dist/prompts/communication-to-intent.system.md`. This explains exactly two
+tests and does not indicate provider or schema failure.
+
+## Execution plan
+
+1. Wait for explicit human approval before editing source.
+2. Correct the relative prompt root by one directory in the owning helper.
+3. Run build and focused communication LLM tests with their mocked provider.
+4. Run the full fresh suite and verify both prompt failures disappear.
+5. Record diff/scope and governance evidence; do not change tests or prompts.
+
+## Actual changes
+
+- The human explicitly approved ticket-033 by replying `kontynuuj`; the
+  interactive implementation may enter `EDIT`.
+- Corrected the prompt traversal by one directory in the owning helper. No
+  prompt content, provider behavior, schema or test changed.
+- Clean build and all three focused communication LLM tests pass, including
+  correction retry and strict require-mode behavior. The isolated full suite
+  no longer contains either prompt-resolution failure.
+- Verified the four sibling implementation commits together in a detached,
+  uncommitted integration worktree: build passes and all 337 runnable tests
+  pass. This is test evidence, not merge approval.
+
+## Blockers
+
+- Tickets 030, 031 and 032 remain independent sibling repairs and are not
+  absorbed into this diff.
+- Four inherited ticket-018/ticket-019 governance errors remain outside scope.
+- Trusted merge approval for the final head SHA is still required.

project/ticket-033/changelog.md

@@ -0,0 +1,18 @@
+# Ticket Changelog (ticket-033)
+
+## [0.1.0] - 2026-08-04
+
+- Initial governance scaffold created.
+- No human participant identity or content was generated.
+- Renumbered the unpublished scaffold from 023 to 033 because remote PR #3
+  already owns ticket-023.
+- Documented the bounded prompt-resolution regression repair.
+
+## [0.2.0] - 2026-08-04
+
+- Corrected communication prompt resolution for the helper's post-refactor
+  directory depth without changing prompt contents or LLM policy.
+- Focused communication LLM tests pass 3/3 and the two targeted full-suite
+  failures are removed.
+- Combined tickets 030-033 implementation verification passes all 337 runnable
+  tests on a clean build.

project/ticket-033/intent.json

@@ -0,0 +1,17 @@
+{
+  "schema": "new-project.intent/v2",
+  "ticket": "ticket-033",
+  "summary": "Restore communication prompt resolution after module split",
+  "workstream": "interfaces",
+  "allowedPaths": [
+    "src/communication/llm/implementation-helpers.ts",
+    "project/ticket-033/**",
+    "TODO.md",
+    "project/TICKETS.md"
+  ],
+  "forbiddenPaths": ["project/ticket-*/user-*.md"],
+  "stacks": ["node"],
+  "dependsOn": [],
+  "conflictsWith": [],
+  "integrationTicket": null
+}

project/ticket-033/preprompt.md

@@ -0,0 +1,8 @@
+# Ticket preprompt
+
+- **Task ID**: ticket-033
+- **Task title**: Restore communication prompt resolution after module split
+- **Created**: 2026-08-04T12:22:18Z
+
+Keep executable implementation outside this governance/evidence directory.
+Read a human-owned user-*.md file only when one exists.

src/communication/llm/implementation-helpers.ts

@@ -297,7 +297,7 @@ export function audit(
 }
 
 export async function readPrompt(): Promise<string> {
-  const promptPath = path.resolve(path.dirname(fileURLToPath(import.meta.url)), '../../../prompts', 'communication-to-intent.system.md');
+  const promptPath = path.resolve(path.dirname(fileURLToPath(import.meta.url)), '../../../../prompts', 'communication-to-intent.system.md');
   if (!(await pathExists(promptPath))) throw new Error(`Prompt not found: ${promptPath}`);
   return fs.readFile(promptPath, 'utf8');
 }