feat(nexus): retry strategies, verify CLI, config threshold, work_proposal, stress bridge (WI-NEXUS-016..020) (#74)

Squash-merge of WI-NEXUS-016..020

Author: tbitcs

.specsmith/ledger-chain.txt

@@ -18,3 +18,8 @@ d0e80ec48ee0a854345237c2fcb8f2ad112ff4f66dc7a6732926017501d85fb4
 7125182a6d402b2e8022fee66cc10950e5734d21e4c40cd1410e9aaca303f5a2
 16eb05be2f953074e4e4c47efb0fd37e9000777db8ddedceee2eca165cf4d925
 01f963eb2078b1815be7cccff4d92dedf3b242c302ceba840f0eb489fe7a4f4a
+0a696105c598f0cd195ef138a2b78ae1e8bc780f58cdea0d0e40d36751281226
+b6a7f5cccf5d0e7064503f161fe685d1108b5541bf2625679ed1a9529147e07b
+b0caf9452cdd3cd154ab6af5d2b8c950a3b8714a5dd9bf7cd54177810e238eac
+32c2742d1f5b332322b25038a5cff4b4e3c25437e3dd16afa8ed24387f6935bd
+1b5b01b80278aabd1ad5ba1599825a7dacd7b20e65ea27dccc98d0a55fdaa84d

.specsmith/requirements.json

@@ -663,5 +663,40 @@
     "description": "A live or honestly-skipped invocation of `scripts/nexus_smoke.py` must be captured under `.specsmith/runs/WI-NEXUS-011/logs.txt` so the project ledger preserves at least one reproducible record of the broker -> preflight -> orchestrator -> vLLM end-to-end path (or a documented reason the live container could not be reached in the current environment).",
     "source": "ARCHITECTURE.md",
     "status": "defined"
+  },
+  {
+    "id": "REQ-096",
+    "title": "Bounded-Retry Harness Must Map Failures to Retry Strategies",
+    "description": "When `execute_with_governance` exhausts its retry budget (REQ-014), it must classify the last executor report against the canonical retry strategy mapping (REQ-028): `narrow_scope`, `expand_scope`, `fix_tests`, `rollback`, or `stop`. The classification must be exposed on `RunResult.strategy` and surfaced in the clarifying question (REQ-063) so the user gets one concrete next-action label rather than only a free-form sentence.",
+    "source": "ARCHITECTURE.md",
+    "status": "defined"
+  },
+  {
+    "id": "REQ-097",
+    "title": "specsmith verify CLI Subcommand",
+    "description": "The Specsmith CLI must expose a `specsmith verify` subcommand that consumes the verification input contract (REQ-027): file diffs, test results, execution logs, and changed files (paths or `--stdin` JSON). The subcommand must emit a JSON object with at least `equilibrium`, `confidence`, `summary`, `files_changed`, `test_results`, and `retry_strategy`. Exit code 0 on equilibrium with confidence ≥ the configured threshold, 2 when retry is recommended, and 3 when stop-and-align is required.",
+    "source": "ARCHITECTURE.md",
+    "status": "defined"
+  },
+  {
+    "id": "REQ-098",
+    "title": "Confidence Threshold Must Be Read From .specsmith/config.yml",
+    "description": "Both `specsmith preflight` and the broker's `run_preflight` helper must consult `.specsmith/config.yml` for the `epistemic.confidence_threshold` value (REQ-058) and use it as the floor for the JSON `confidence_target` field whenever it is greater than the heuristic default. When the config file is absent or unparseable, the existing heuristic defaults must continue to apply.",
+    "source": ".specsmith/config.yml, ARCHITECTURE.md",
+    "status": "defined"
+  },
+  {
+    "id": "REQ-099",
+    "title": "Accepted Preflight Must Record a Distinct work_proposal Event",
+    "description": "When `specsmith preflight` produces an `accepted` decision and assigns a brand-new `work_item_id`, the CLI must append a `work_proposal` ledger event in addition to the existing `preflight` event (REQ-044). The `work_proposal` entry must reference REQ-044 and REQ-085, include the `work_item_id` and matched `requirement_ids`, and must NOT be emitted when the underlying `work_item_id` already appears in `LEDGER.md` (no duplicate proposals).",
+    "source": "ARCHITECTURE.md",
+    "status": "defined"
+  },
+  {
+    "id": "REQ-100",
+    "title": "Broker Scope Inference May Surface Stress-Test Critical Failures",
+    "description": "When the user passes `--stress` to `specsmith preflight` and the matched requirements set is non-empty, the CLI must invoke the existing AEE `StressTester` against those belief artifacts and surface any critical failures in the JSON payload as a `stress_warnings` list. The narration (verbose mode) must include a one-sentence plain-English warning when at least one critical failure is found. The flag must default off so unrelated tests continue to pass.",
+    "source": "ARCHITECTURE.md",
+    "status": "defined"
   }
 ]

.specsmith/runs/WI-NEXUS-015/pr-body.md

@@ -0,0 +1,59 @@
+# feat(nexus): TaskResult, preflight exit codes, ledger event, /why post-run, smoke evidence
+
+Tightens the **Nexus** ↔ **Specsmith** contract that landed in PR #72. Five
+follow-up work items, all governed by Specsmith and verified by pytest.
+**Suite: 247 passing, 1 skipped (live l1-nexus integration test).**
+
+## Work items in this PR
+
+- **WI-NEXUS-011 (REQ-095)** — Captured live `l1-nexus` smoke evidence at
+  `.specsmith/runs/WI-NEXUS-011/logs.txt`. The smoke script ran offline and
+  returned a structured `ok=false` transport error; the log includes a
+  reproducible note describing how to re-run it against a live container.
+- **WI-NEXUS-012 (REQ-091)** — `orchestrator.run_task` now returns a
+  `TaskResult` dataclass (`equilibrium`, `confidence`, `summary`,
+  `files_changed`, `test_results`). The Nexus REPL's bounded-retry harness
+  consumes it directly instead of synthesizing equilibrium from
+  `bool(summary)`. Adds a tolerant parser for the existing Nexus output
+  contract (Plan/Commands/Files changed/Diff/Test results/Next action).
+- **WI-NEXUS-013 (REQ-094)** — Nexus REPL emits a `[/why]` post-run
+  governance block when `verbose_governance` is on, listing the assigned
+  `work_item_id`, matched `requirement_ids`/`test_case_ids`, post-run
+  `confidence`, and harness `equilibrium`.
+- **WI-NEXUS-014 (REQ-092)** — `specsmith preflight` exits `0` for
+  `accepted`, `2` for `needs_clarification`, and `3` for
+  `blocked`/`rejected`. The JSON payload continues to print on stdout for
+  every exit code so CI pipelines can branch on intent without re-parsing.
+- **WI-NEXUS-015 (REQ-093)** — Every accepted `specsmith preflight` invocation
+  appends a `preflight` ledger event tagged with `REQ-085` plus the matched
+  `requirement_ids`, recording the utterance, assigned `work_item_id`, and
+  `confidence_target`. Non-accepted decisions never touch the ledger.
+
+## Verification
+
+- `py scripts/sync_governance_state.py` → 95 requirements / 95 test cases.
+- `py -m pytest -q` → **247 passed, 1 skipped** (≈17s; the skip is the
+  `NEXUS_LIVE=1`-gated integration test).
+- Smoke evidence: `.specsmith/runs/WI-NEXUS-011/logs.txt`.
+- Cumulative diff + final pytest log: `.specsmith/runs/WI-NEXUS-015/`.
+- Five new ledger entries chained for WI-NEXUS-011..015.
+
+## Notes for reviewers
+
+- The post-run `[/why]` block is gated entirely behind the existing `/why`
+  toggle; default REPL behavior remains plain English with no governance
+  identifiers leaking to the user.
+- The orchestrator's heuristic confidence (0.85 on full contract, 0.4
+  partial) is documented as a placeholder for a real verifier signal; the
+  retry harness already honors whatever value the executor returns.
+- The preflight ledger writer is best-effort — ledger errors never block
+  the CLI from emitting its JSON or returning its exit code.
+
+---
+
+🤖 Generated with [Warp](https://app.warp.dev) — agent conversation:
+[link](https://app.warp.dev/conversation/6f8aa790-049b-4ddf-9c52-4840728faee5)
+
+Plan artifact: [Warp Agent Implementation Plan](https://app.warp.dev/drive/notebook/rfCwIZUgJPCakjJ2S552DX)
+
+Co-Authored-By: Oz <oz-agent@warp.dev>