loki-mode 9.77.0 → 9.80.1

package/docs/v10/ENGINE.md

@@ -0,0 +1,272 @@
+
+
+## 23:00Z cut (E-31..E-40)
+
+Every claim below comes from reading main at 84c22568. I edited no files.
+
+**What I found that shapes this cut**
+
+1. **The engine cannot run from the npm package today.**
+   - Root `package.json` `files` ships `loki-ts/dist/` and no `loki-ts/src`. Every source-relative path is therefore dead in an install. These break:
+     - `machine.ts:44-55`: `optional()` checks whether `import.meta.dir/stages/<name>.ts` exists. In a bundle that file is never there, so every stage emits `stage.skipped` "module not present".
+     - `engine10/cli.ts:54`: the default loader uses a non-literal `import(spec)`.
+     - `output.ts:132`: dynamic import of the eta module.
+     - `session.ts:137`: spawns `[process.execPath, [import.meta.path, "--engine10-session-child"]]`. In a bundle, `import.meta.path` is `dist/loki.js`.
+     - `fetch_issue.ts:32` and `stages/pr.ts:49`: paths built with `new URL("../../../...", import.meta.url)`, which resolve above the package root from dist.
+   - `util/paths.ts` already finds `REPO_ROOT` by walking up to `VERSION` + `autonomy/run.sh`, and works from dist. The fix is to reuse it.
+2. **The flip breaks two merged tests.**
+   - `tests/test-engine10-dispatch.sh` and `tests/test-engine10-legacy-contract.sh` both assert that an unset `LOKI_ENGINE` routes to legacy.
+   - The E-30 alias assertions switch from SKIP to live only when bin/loki matches `^[[:space:]]*legacy\)`.
+3. **The gate chain has a hole.** No row runs the v10 arm over the 29 tasks. EV-4 is raw claude and EV-5 is legacy. E-38 closes this, and it runs the dist build so the gate measures what actually ships.
+4. **Hard limits from existing code:**
+   - `loki-ts/src/runner/types.ts:8` has no opencode provider, and `providers.ts` is on the no-edit list. So the only honest opencode path is a clear refusal.
+   - `autonomy/loki` is also no-edit, so the deprecation note cannot live in legacy help text.
+   - `supervisor.ts` has no `main` yet. Its wiring arrives with E-14, so preflight wiring waits for E-14.
+5. **E-TG's file set is not recorded anywhere.** PROGRESS.md:1255 names it only as a follow-up to E-07 and E-09. E-32 touches `session.ts`, so it depends on E-TG merging.
+
+**How the file sets stay separate**
+- **In-flight rows:** no file below appears in any ready, building or review row (E-14..E-29, EV-4, EV-5, EV-6, E-TG).
+- **EV-6:** it keeps `eval/loki10/summarize` and `docs/v10/METRICS.md`. E-33 builds a new generator that EV-6 runs, so the two sets stay apart.
+- **One planned exception:**
+  - E-31 changes one marked line in `README.md` and one in `docs/v10/GUIDE.md` (the opt-in line that E-34 writes).
+  - E-31 depends on E-34, so the two are never in flight together.
+  - The Captain applies that line change in the E-31 merge commit.
+
+### E-31: Default flip to v10, with a `LOKI_ENGINE=legacy` override and the `loki legacy` alias
+- **Files:**
+  - `bin/loki`
+  - `loki-ts/src/engine10/cli.ts` (USAGE text only)
+  - `tests/test-engine10-default.sh` (new)
+  - `tests/test-engine10-dispatch.sh`
+  - `tests/test-engine10-legacy-contract.sh`
+  - After E-34 merges: the one marked opt-in line in `README.md` and in `docs/v10/GUIDE.md`
+- **Tier:** HIGH.
+- **Gating:**
+  - Build on a branch now.
+  - After review, hold the row at `approved`. Merge only when the E-33 marker in `docs/v10/METRICS.md` reads `<!-- loki10-gate: met ... -->`.
+  - If the marker does not read met by 03:00Z, the row goes to `parked` and v10.0.0 ships with legacy as the default (D29).
+  - The test enforces the gate mechanically: the build goes red if the switch reads v10 while the marker does not read met.
+- **Contents:**
+  - **One switch.** A single line in bin/loki, `_loki_engine_default="v10"`, and the v10 block tests `"${LOKI_ENGINE:-$_loki_engine_default}"`.
+  - **Override values.**
+    - `LOKI_ENGINE=legacy` runs today's route exactly.
+    - Any other non-empty value except `v10` exits 2 with a one-line message, so a typo such as `legasy` fails instead of silently picking an engine.
+  - **Alias.** A `case "${1:-}" in legacy) shift; export LOKI_ENGINE=legacy ;; esac` arm runs before the v10 block. The export keeps nested calls on legacy.
+    - For `loki legacy`, `loki legacy --help`, `-h` or `help`, the arm first prints `autonomy/lib/engine10-legacy-notice.txt` to stderr if that file exists (E-35 owns the text).
+  - **Precedence.** `LOKI_LEGACY_BASH=1` still wins over everything.
+  - **No bun.**
+    - A default v10 selection without bun prints one notice line ("v10 needs bun; running the legacy engine. Set LOKI_ENGINE=legacy to silence") and falls through to legacy.
+    - An explicit `LOKI_ENGINE=v10` without bun still exits 1, as today.
+  - **Help after the flip.** `loki`, `loki --help` and `loki -h` go to the engine10 USAGE. USAGE ends with `loki legacy <args>  run the previous engine (deprecated)`. `--version` stays on the legacy route.
+  - **Commands that change meaning.** `status`, `verify` and `dashboard` now mean the v10 commands. The legacy ones are reached through `loki legacy status` and so on.
+- **Wall:** `bash tests/test-engine10-default.sh && bash tests/test-engine10-legacy-contract.sh && bash tests/test-engine10-dispatch.sh`. The output must show 0 failed for all three, and "0 skipped" on the legacy-contract Results line.
+- **Red:** the default test fails because nothing flips the default, and the legacy contract reports 20 SKIP lines.
+- **Green:**
+  - With `LOKI_ENGINE` unset, a task routes to engine10.
+  - `LOKI_ENGINE=legacy` and `loki legacy <args>` each match all 20 golden routes.
+  - `LOKI_LEGACY_BASH` wins.
+  - With no bun, the default falls back with a notice and explicit v10 exits 1.
+  - `LOKI_ENGINE=legasy` exits 2.
+  - `grep -c '_loki_engine_default=' bin/loki` equals 1.
+  - Planting a METRICS.md without the met marker turns the default test red.
+- **Deps:** E-30, E-32, E-33, E-34, E-35, E-38, EV-6.
+
+### E-32: Packaging, so the engine runs from dist and the npm package
+- **Files:**
+  - `loki-ts/src/engine10/registry.ts` (new)
+  - `loki-ts/src/cli.ts` (inside the existing engine10 arm only)
+  - `loki-ts/src/engine10/machine.ts`
+  - `loki-ts/src/engine10/session.ts`
+  - `loki-ts/src/engine10/output.ts`
+  - `loki-ts/src/engine10/fetch_issue.ts`
+  - `loki-ts/src/engine10/stages/pr.ts`
+  - `loki-ts/tests/engine10/registry.test.ts` (new)
+  - `tests/test-engine10-dist.sh` (new)
+- **Tier:** HIGH. It changes how provider sessions spawn and the path to the credentialed push child.
+- **Contents:**
+  - **Registry.** `registry.ts` maps each stage and route module name to a literal lazy import, for example `wall: () => import("./stages/wall.ts")`. Only modules present on disk at build time are listed. Bun bundles literal imports, and loading stays lazy.
+  - **Loaders.**
+    - `machine.ts` `defaultLoader` and `output.ts`'s eta load go through the registry.
+    - The `src/cli.ts` arm passes the registry loader to `runEngine10(rest, load)`. The string `engine10` stays inside that one arm, which `cli.test.ts` asserts.
+    - `optional()` stays as the source fallback.
+  - **Session child.** `session.ts` spawns the child as `[process.execPath, <bundle or src cli entry>, "engine10", "session"]`, using the `session` route that already exists in the TABLE, instead of `import.meta.path`.
+  - **Paths.** `fetch_issue.ts` and `pr.ts` resolve `autonomy/issue-providers.sh` and `autonomy/lib/engine10-push.sh` from `REPO_ROOT` (`util/paths.ts`).
+  - **Registration guard.** `registry.test.ts` fails when a module that the stage table or cli TABLE names exists on disk but is not registered. Slices that land after E-32 (wave 2 still building, E-24..E-29) are each one registry line, which the Captain adds at merge.
+  - **No writes to `loki-ts/dist/`** (section 14).
+- **The dist test:**
+  1. `npm pack` into a temp dir and extract it. The result has no `loki-ts/src`.
+  2. `bun build` the current src into `<extract>/loki-ts/dist/loki.js`, simulating the Captain's rebuild.
+  3. Run `<extract>/bin/loki` with `LOKI_ENGINE=v10 LOKI_E10_INVOKER=cli LOKI_CLAUDE_CLI=<E-14 stub>` on a copy of the E-14 fixture with `--no-pr`.
+  4. Separately, `npm pack --dry-run --json` must list `autonomy/lib/engine10-push.sh`, `autonomy/receipt_jwt.py` and `loki-ts/dist/loki.js`.
+- **Wall:** `bash tests/test-engine10-dist.sh && (cd loki-ts && bun test tests/engine10/registry.test.ts)`.
+- **Red:** the same script on current main shows every stage `stage.skipped` with "module not present", and the session child never starts.
+- **Green:**
+  - `events.jsonl` from the extracted package holds `stage.completed` for intake, implement, verify and seal.
+  - There are 0 "module not present" skips for landed modules.
+  - `receipt.json`, `.loki/engine.json` and an efficiency record exist.
+  - The dry-run list contains the three files.
+- **Deps:** E-14, E-TG.
+
+### E-33: EV-6 gate report generator
+- **Files:** `eval/loki10/gate_report.py` (new), `eval/loki10/test-gate-report.sh` (new), `eval/loki10/fixtures/gate/` (new subdir).
+- **Tier:** HIGH. Its marker controls E-31.
+- **Contents:**
+  - **Invocation:** `gate_report.py <results.jsonl> --metrics <path> --changelog <path>`. It reuses `harness.summarize_rows`.
+  - **Output:**
+    - a per-arm table (completed/evaluated, rate, p50 and p90 time to PR, cost per completed task, runs with measured cost);
+    - four gate lines, each showing the measured number, the threshold and MET or MISSED;
+    - every miss, as task id plus status or reason;
+    - null cost rendered as "not measured".
+  - **`met` requires all of:**
+    - at least 25 evaluated tasks on each arm;
+    - the same `model` on every row;
+    - v10 completion at or above raw-claude;
+    - v10 p50 of 300s or less and p90 of 600s or less;
+    - measured cost on both arms, with v10 cost per completed task at or below raw-claude. Null cost counts as MISSED.
+  - **Markers:** it writes `<!-- loki10-gate: met|missed n=<N> results_sha256=<sha> -->` and replaces the block between `<!-- loki10-gate:begin -->` and `<!-- loki10-gate:end -->` in both target files, so rerunning is idempotent.
+  - **Who writes the real files:** EV-6 runs this on real results and commits METRICS.md and the CHANGELOG block. This slice never touches `summarize` or METRICS.md.
+- **Wall:** `bash eval/loki10/test-gate-report.sh`.
+- **Red:** the file is missing.
+- **Green:**
+  - An all-met fixture writes the met marker.
+  - A v10 p90 of 660s gives MISSED and lists the slow tasks.
+  - Null v10 cost gives MISSED with "not measured".
+  - 24 evaluated tasks gives MISSED with "fewer than 25".
+  - A model mismatch gives MISSED.
+  - A second run leaves exactly one block.
+- **Deps:** EV-1.
+
+### E-34: v10 user docs, marked opt-in until the flip
+- **Files:** `README.md` (new quickstart section), `docs/v10/GUIDE.md` (new), `tests/test-engine10-docs.sh` (new).
+- **Tier:** LOW.
+- **Contents:**
+  - **Quickstart:** `LOKI_ENGINE=v10 loki "<task>"`, issue refs, and `--no-pr`, `--deep`, `--provider`, `--resume`.
+  - **The 5-line summary,** explained line by line.
+  - **NOT PROVEN:** what it means, including the deep checks that are always listed and "not measured" cost.
+  - **Receipts:** `receipt.json` and `receipt.md`, `loki verify`, and UNSIGNED.
+  - **Provider table:** claude cost is measured. For codex, cline and aider, cost is not measured and kill blocking is not enforced. opencode is not supported and is refused.
+  - **After the flip:** `status`, `verify` and `dashboard` change meaning. `loki legacy <args>` and `LOKI_ENGINE=legacy` reach the old engine.
+  - **The opt-in line:** one line in each file carries `<!-- loki10-default -->` and reads "Opt-in: set LOKI_ENGINE=v10". E-31 changes that line.
+- **Wall:** `bash tests/test-engine10-docs.sh`.
+- **Red:** GUIDE.md is missing.
+- **Green:**
+  - Every command in the guide appears in the engine10 USAGE.
+  - The marked line's wording matches bin/loki's current default.
+  - README links GUIDE.md.
+  - The docs contain no em or en dashes.
+- **Deps:** E-13, E-22.
+
+### E-35: Legacy deprecation note, with no removal
+- **Files:** `autonomy/lib/engine10-legacy-notice.txt` (new), `CHANGELOG.md` (a "Deprecated" entry for v10.0.0), `tests/test-engine10-legacy-notice.sh` (new).
+- **Tier:** LOW.
+- **Contents:**
+  - The notice runs three lines or fewer. It says the legacy engine is deprecated in v10.0.0 and still fully supported, that nothing is removed in this release, that `LOKI_ENGINE=legacy` pins it, and where the guide lives.
+  - The notice avoids `autonomy/loki`, which is on the no-edit list. E-31's `legacy)` arm prints the file for help invocations only.
+  - The CHANGELOG entry states the same facts and says no removal date is set.
+- **Wall:** `bash tests/test-engine10-legacy-notice.sh`.
+- **Red:** the file is missing.
+- **Green:**
+  - The notice exists and has 3 lines or fewer.
+  - It contains "LOKI_ENGINE=legacy" and does not contain "removed in".
+  - The CHANGELOG has a Deprecated entry that names `loki legacy`.
+  - Once E-31 is present, `loki legacy --help` stderr contains the first notice line. Until then that assertion prints SKIP.
+- **Deps:** none (the file must exist before E-31 merges).
+
+### E-36: First-run preflight with clear errors
+- **Files:** `loki-ts/src/engine10/preflight.ts` (new), `loki-ts/tests/engine10/preflight.test.ts` (new), `loki-ts/src/engine10/supervisor.ts` (one call at the start of `main`, before the marker or branch).
+- **Tier:** MEDIUM.
+- **Checks:** each failure prints one line with a fix hint and exits 2 before any branch or event log exists.
+  - **Not a repo:** the current directory is not a git work tree.
+  - **Git identity:** `user.name` or `user.email` is unset, which would make the commit fail.
+  - **Provider CLI:** the provider's CLI is not on PATH (`LOKI_CLAUDE_CLI`, `LOKI_CODEX_CLI` and so on are honored).
+  - **Unsupported provider:** `--provider opencode`, or any other unknown name, gets: "the v10 engine has no opencode invoker yet; use LOKI_ENGINE=legacy loki start --provider opencode".
+  - **Host guard:** an inherited `LOKI_HOST_GUARD=1` with a non-claude provider would throw at `providers.ts:63`, so preflight explains it first.
+  - **GitHub CLI:** when origin is on github.com and `--no-pr` is absent, missing `gh` or a failing `gh auth status` gets a hint to use `--no-pr`. A non-GitHub origin or `--no-pr` skips the check, so the eval's local bare remote is unaffected.
+- **Warnings (not fatal):** `python3 -I` without `cryptography`, or no signing key, prints "receipts will be UNSIGNED".
+- **bun:** bin/loki already checks for bun.
+- **Wall:** `cd loki-ts && bun test tests/engine10/preflight.test.ts`.
+- **Red:** the file is missing.
+- **Green:**
+  - Each fatal case gives exit 2, its exact line, and no `.loki/runs` dir.
+  - Missing cryptography warns and continues.
+  - `--no-pr` skips the gh check.
+- **Deps:** E-14.
+
+### E-37: Non-claude provider paths through session.ts, with honest cost
+- **Files:** `loki-ts/tests/engine10/providers.test.ts` (new), `loki-ts/tests/engine10/fixtures/providers/` (new: stub codex, cline and aider CLIs).
+- **Tier:** MEDIUM.
+- **Contents:**
+  - Runs the supervisor end to end with `--provider codex`, `cline` and `aider` through their stubs (`LOKI_CODEX_CLI`, `LOKI_CLINE_CLI`, `LOKI_AIDER_CLI`) and `--no-pr`.
+  - This slice has no source files. Any red it finds becomes its own fix slice.
+- **Wall:** `cd loki-ts && bun test tests/engine10/providers.test.ts`.
+- **Red:** the file is missing.
+- **Green, for each non-claude provider:**
+  - The run reaches seal.
+  - The `cost` event has `usd: null` with tokens kept.
+  - `cost-summary.py --json` reports `fully_measured` false.
+  - The Cost line reads "not measured" and never "$0.00".
+  - The receipt's `not_proven` holds "kill blocking not enforced".
+  - With `LOKI_MODEL_OVERRIDE` set, it also holds "model override not applied".
+  - The session child env has no `LOKI_HOST_GUARD`.
+  - `--provider opencode` exits 2 with the E-36 line.
+- **Deps:** E-14, E-36.
+
+### E-38: v10 arm gate run, with the legacy arm pinned
+- **Files:** `eval/loki10/harness.py` (the legacy arm sets `env["LOKI_ENGINE"]="legacy"`), `eval/loki10/test-harness.sh` (one new leg).
+- **Tier:** HIGH.
+- **Contents:**
+  - Pinning the legacy arm keeps EV-5 results unchanged after the flip.
+  - The v10 arm runs `--all` over the 29 tasks using the dist build from E-32 (`LOKI_TS_ENTRY` pointing at the rebuilt bundle), so the gate measures what ships.
+  - Results stay under the gitignored `eval/loki10/results/`.
+- **Wall:** `bash eval/loki10/test-harness.sh && eval/loki10/run.sh --arm v10 --all`. The results must hold 29 rows, none `arm_unavailable`, and each row must carry `model`, `time_to_pr_s` and `cost_usd` (a number or null).
+- **Red:** the new leg fails because the legacy arm env has no `LOKI_ENGINE`, and no v10 results exist.
+- **Green:** the leg asserts `LOKI_ENGINE=legacy` on the legacy arm and `v10` on the v10 arm, and the 29-row v10 results file exists.
+- **Deps:** EV-1, E-32, E-36.
+
+### E-39: Interrupt and resume end to end
+- **Files:** `loki-ts/tests/engine10/resume_e2e.test.ts` (new), `loki-ts/tests/engine10/fixtures/resume/` (new).
+- **Tier:** MEDIUM.
+- **Wall:** `cd loki-ts && bun test tests/engine10/resume_e2e.test.ts`.
+- **Red:** the file is missing.
+- **Green:**
+  - SIGINT during a sleeping implement stub exits 130. The session process group is gone and there is no `run.completed`.
+  - `loki --resume <id>` finishes through seal on the same run id.
+  - `seq` keeps increasing, and there is no second `stage.started` for intake.
+  - `.loki/engine.json` still names the same run.
+  - Changing the origin between runs refuses the push.
+- **Deps:** E-14.
+
+### E-40: Live PR smoke on a sandbox GitHub repo
+- **Files:** `tests/test-engine10-live-pr.sh` (new).
+- **Tier:** HIGH (credentialed).
+- **Contents:**
+  - Runs one real small task from the E-32 dist layout against `LOKI_E10_LIVE_REPO`.
+  - Without the variable it prints SKIP and exits 0. SKIP never counts as done; the Wall command sets the variable.
+- **Wall:** `LOKI_E10_LIVE_REPO=<owner>/<sandbox> bash tests/test-engine10-live-pr.sh`. The output must show:
+  - a PR URL;
+  - `gh pr view --json isDraft` agreeing with the verdict;
+  - the `loki/deep-verify` status context through `gh api`;
+  - `loki verify <run>` exiting 0, or UNSIGNED stated;
+  - a second `--resume` reporting `existing: true` with the same URL.
+- **Red:** the file is missing.
+- **Green:** all five assertions hold against the sandbox repo.
+- **Deps:** E-23, E-32, E-36.
+
+### BOARD rows
+
+The notes start with the requested "Source: ENGINE.md 23:00Z cut." and then add dependencies, following the 21:50Z rows. If you want the notes to be only that sentence, drop the dependency part.
+
+| id | title | files | tier | wall check | status | notes |
+|---|---|---|---|---|---|---|
+| E-31 | Default flip to v10 with LOKI_ENGINE=legacy override and the loki legacy alias (merge held until the gate marker reads met) | bin/loki, loki-ts/src/engine10/cli.ts (USAGE only), tests/test-engine10-default.sh, tests/test-engine10-dispatch.sh, tests/test-engine10-legacy-contract.sh, README.md and docs/v10/GUIDE.md (marked opt-in line only, after E-34) | HIGH | bash tests/test-engine10-default.sh && bash tests/test-engine10-legacy-contract.sh && bash tests/test-engine10-dispatch.sh (0 failed; legacy contract 0 skipped) | ready@2026-09-27T23:00Z | Source: ENGINE.md 23:00Z cut. Held at approved until the METRICS.md loki10-gate marker reads met; parked if not met by 03:00Z. Depends on E-30, E-32, E-33, E-34, E-35, E-38, EV-6. |
+| E-32 | Packaging: engine runs from dist and the npm package (static registry, session child via cli route, REPO_ROOT paths) | loki-ts/src/engine10/registry.ts, loki-ts/src/cli.ts (engine10 arm only), loki-ts/src/engine10/machine.ts, loki-ts/src/engine10/session.ts, loki-ts/src/engine10/output.ts, loki-ts/src/engine10/fetch_issue.ts, loki-ts/src/engine10/stages/pr.ts, loki-ts/tests/engine10/registry.test.ts, tests/test-engine10-dist.sh | HIGH | bash tests/test-engine10-dist.sh && (cd loki-ts && bun test tests/engine10/registry.test.ts) | ready@2026-09-27T23:00Z | Source: ENGINE.md 23:00Z cut. Later engine slices add one registry line at merge. Depends on E-14, E-TG. |
+| E-33 | Gate report generator: METRICS.md and CHANGELOG blocks with every gate number, misses and the loki10-gate marker | eval/loki10/gate_report.py, eval/loki10/test-gate-report.sh, eval/loki10/fixtures/gate/ | HIGH | bash eval/loki10/test-gate-report.sh | ready@2026-09-27T23:00Z | Source: ENGINE.md 23:00Z cut. EV-6 runs it on real results and commits METRICS.md. Depends on EV-1. |
+| E-34 | v10 user docs: README quickstart and docs/v10/GUIDE.md (5-line summary, NOT PROVEN, receipts), opt-in until the flip | README.md, docs/v10/GUIDE.md, tests/test-engine10-docs.sh | LOW | bash tests/test-engine10-docs.sh | ready@2026-09-27T23:00Z | Source: ENGINE.md 23:00Z cut. Depends on E-13, E-22. |
+| E-35 | Legacy engine deprecation note for loki legacy help and CHANGELOG (no removal) | autonomy/lib/engine10-legacy-notice.txt, CHANGELOG.md, tests/test-engine10-legacy-notice.sh | LOW | bash tests/test-engine10-legacy-notice.sh | ready@2026-09-27T23:00Z | Source: ENGINE.md 23:00Z cut. Depends on none; lands before E-31. |
+| E-36 | First-run preflight: repo, git identity, provider CLI, opencode refusal, host guard, gh auth, UNSIGNED warning | loki-ts/src/engine10/preflight.ts, loki-ts/tests/engine10/preflight.test.ts, loki-ts/src/engine10/supervisor.ts (one call in main) | MEDIUM | cd loki-ts && bun test tests/engine10/preflight.test.ts | ready@2026-09-27T23:00Z | Source: ENGINE.md 23:00Z cut. Depends on E-14. |
+| E-37 | Non-claude provider paths (codex, cline, aider) through session.ts with cost not measured and NOT PROVEN entries | loki-ts/tests/engine10/providers.test.ts, loki-ts/tests/engine10/fixtures/providers/ | MEDIUM | cd loki-ts && bun test tests/engine10/providers.test.ts | ready@2026-09-27T23:00Z | Source: ENGINE.md 23:00Z cut. Test only; a red becomes its own slice. Depends on E-14, E-36. |
+| E-38 | v10 arm gate run over the 29 tasks from dist; legacy arm pinned to LOKI_ENGINE=legacy | eval/loki10/harness.py (legacy arm env), eval/loki10/test-harness.sh (one leg) | HIGH | bash eval/loki10/test-harness.sh && eval/loki10/run.sh --arm v10 --all (29 rows, none arm_unavailable) | ready@2026-09-27T23:00Z | Source: ENGINE.md 23:00Z cut. The missing third arm for the gate. Depends on EV-1, E-32, E-36. |
+| E-39 | Interrupt and resume end to end (SIGINT, no orphan, resume to seal on the same run) | loki-ts/tests/engine10/resume_e2e.test.ts, loki-ts/tests/engine10/fixtures/resume/ | MEDIUM | cd loki-ts && bun test tests/engine10/resume_e2e.test.ts | ready@2026-09-27T23:00Z | Source: ENGINE.md 23:00Z cut. Depends on E-14. |
+| E-40 | Live PR smoke on a sandbox GitHub repo from the dist layout | tests/test-engine10-live-pr.sh | HIGH | LOKI_E10_LIVE_REPO=<owner>/<sandbox> bash tests/test-engine10-live-pr.sh | ready@2026-09-27T23:00Z | Source: ENGINE.md 23:00Z cut. SKIP without the variable does not count as done. Depends on E-23, E-32, E-36. |
+
+**Critical path to the flip:** E-14 and E-TG → E-32 → E-36 → E-38 → EV-6 (runs E-33) → E-31. E-34 and E-35 must merge before E-31. E-37, E-39 and E-40 can run in parallel once their dependencies land.

package/docs/v10/METRICS.md

@@ -217,3 +217,29 @@ remains the only cap on this step. Guarded by
 ## Over-budget stops
 
 - 2026-09-27T21:33Z: batch 8 workflow wf_ff8d537c-40e stopped. Its last two reviewers (S-191, S-195) ran about 55 minutes against the 30-minute reviewer budget with no verdict (pulse AGENT_OVER_BUDGET at 21:31Z: "S-191 building MEDIUM (55.9 min, budget 30 min)"). Re-dispatched as two standalone reviewers with 20 and 25 minute scopes. Load average was 15 to 17 during the stall.
+
+## Loki 10 eval: first real arm numbers (2026-09-27, smoke, not the gate)
+
+Raw `claude -p` arm (EV-4), model claude-opus-5-5, 900s cap, 3 public tasks, harness from the EV-1 branch with EV-3 isolation (auth via keychain access token; no credential in any result file):
+
+| task | completed | time to PR | cost (provider-reported) |
+|---|---|---|---|
+| pub-more-itertools-1192 | yes | 20s | $0.1266 |
+| pub-click-2877 | yes | 34s | $0.1861 |
+| pub-humanize-152 | yes | 45s | $0.2704 |
+
+`summarize --markdown`: 3/3 completed, p50 34s, p90 45s, $0.1944 per completed task, 0 capped. This is a 3-task smoke, not the 29-task gate. It sets the bar the v10 fast lane must meet on completion and cost per completed task (D29), which is far tighter than the 5-minute p50 target.
+
+## Loki 10 eval: raw claude -p arm, full 29 tasks (2026-09-27 23:05-23:16Z)
+
+Model claude-opus-5-5, harness 84c22568, 900s cap, --parallel 3, EV-3 isolation. Results: ~/loki-ci-logs/eval-raw-claude-20260927T231653Z/.
+
+| Arm | Completed | Rate | p50 time to PR | p90 time to PR | Cost per completed | Cost measured | Capped |
+|---|---|---|---|---|---|---|---|
+| raw-claude | 27/29 | 93.1% | 39s | 68s | $0.2363 | 29/29 | 0 |
+
+Misses: pub-click-3059 (hidden tests failed: wrong metavar bracketing), pub-humanize-174 (hidden tests failed: rounding boundary cases). Total provider-reported spend $6.38.
+
+## Loki 10 engine: first real run (E-14 smoke, not the gate)
+
+pub-more-itertools-1192 through the E-14 glue entry, real claude, --no-pr: 30s wall (intake 0s, plan 16s, implement 12s, verify 3s, commit and seal under 1s), hidden test passes, $0.4486 over 2 sessions (plan $0.2264, implement $0.2223), verdict PARTIAL only because ruff is not installed. Raw claude on the same task: 15s, $0.1290. The planner session costs as much as the implement session.

package/docs/v10/PROGRESS.md

@@ -1231,3 +1231,48 @@ sample.
 - INCIDENT: an agent other than the Chief of Staff committed and pushed to main. Commit 28926937 "docs(v10): cite ARCHITECT-CUTS.md, retract unevidenced green claims" (author asklokesh, 14:35:11 local) landed on origin/main (`git ls-remote origin refs/heads/main` = 28926937...). Its content is exactly the Chief of Staff's staged, uncommitted working-tree change, so nothing wrong landed, but builders are forbidden to push or edit BOARD.md, and it broke the release freeze on a75c8b98. The likely cause is a batch-4 worker acting on the relayed founder message. The PreToolUse guard (.claude/settings.local.json in the main checkout) may not apply to agents running inside .claude/worktrees; guard slice S-152 follows.
 - Consequence: release 5 moves from a75c8b98 to 28926937 (docs-only on top); it waits for Tests on that SHA, then bumps through release.sh --bump-only.
 - Batch 4: 13 LOW slices show over budget at 17 min, several of whose builders were diverted by the relayed message; they are collected when the workflow returns, and diverted ones are re-dispatched.
+
+## 2026-09-27T21:47Z: Loki 10 engine P0 started (D29)
+- Leader lock held by this session (`cat .loki/v10-leader` = 74619, the running claude process).
+- Orphan PIDs 22960 and 22986 stopped (SIGTERM ignored, SIGKILL by PID); `ps -o pid= -p 22960` empty.
+- Backlog paused: 22 BOARD rows parked, batch 10 builders' commits kept on their branches.
+- v9.77.0 tagged 21:46:19 (`git ls-remote origin refs/tags/v9.77.0^{}` = b473123d); train cadence continues.
+- In flight: stage-time measurement of augmentiq #52 plus 3 recent runs; eval harness runner (EV-1); eval task curation, 25+ tasks with hidden tests (EV-2); ORPHAN guard (E-00).
+- ETA for v10.0.0: the engine design lands first; the gate decision is at 03:00 UTC (23:00 ET).
+- Top blocker: none yet; the measured stage table decides the design.
+
+## 2026-09-27T22:08Z: Loki 10 engine progress
+- ENGINE.md on main (`git ls-remote origin refs/heads/main` = a4ea867b): measured stage table (docs/v10/ENGINE-MEASURE.md), architecture, 30 slices E-01..E-30.
+- Slices in flight: E-01, E-05, E-06 (phase A, wf_f5dde039-0c5); E-11 shell half, E-12, E-30 (wf_f70c455a-69e); E-00 ORPHAN guard built (5e52dd64), in review.
+- Eval: EV-1 runner built (3cf723be, `bash eval/loki10/test-harness.sh` 34 passed), in adversarial review. EV-2 has 27 tasks (augmentiq 1, public 12, quickstart 14; f9304d1a), under full red/green re-verification. EV-3 arm isolation (clean claude config) in flight.
+- Eval numbers so far: none (no arm has run yet).
+- ETA for v10.0.0: thin path end to end targeted for about 01:00 UTC; gate decision at 03:00 UTC.
+- Top blocker: E-01 (shared types) gates phase B; augmentiq has only 3 issues, so the augmentiq share of the eval is 1 task, not 5.
+
+## 2026-09-27T22:57Z: Loki 10 engine progress
+- On main (`git ls-remote origin refs/heads/main` = 694866fc): engine wave 1 E-01..E-13 except E-14, plus E-11 (both halves), E-16, E-20; the eval harness EV-1 with EV-3 isolation and the 29 EV-2 tasks (EV-7 nonce fixes). tests/engine10 plus spawn guard 185 pass, 0 fail; eval/loki10/test-harness.sh 77 passed.
+- Eval baseline: a no-op arm over all 29 tasks graded every task not completed, none task_invalid (108s). No real arm numbers yet: EV-4 (raw claude -p, 3 tasks) and EV-5 (legacy, 1 task) are running.
+- In flight: E-14 end to end plus the first real v10 run on pub-more-itertools-1192; reworks E-15, E-17, E-18, E-21, E-22; E-19, E-23, E-TG.
+- Incidents fixed forward: two orphan loki test runs over 24 hours (stopped by PID; E-00 guard merged); main CI red on Coverage from engine spawns without env (72423684, 19128f6b).
+- ETA for v10.0.0: first real v10 task within the next hour; gate decision at 03:00 UTC.
+- Top blocker: E-14 integration (first time the modules run together).
+
+## 2026-09-27T23:03Z: first real eval numbers
+- Raw claude -p arm, 3 public tasks: 3/3 completed, p50 34s, p90 45s, $0.1944 per completed task (docs/v10/METRICS.md). Legacy arm and v10 arm not yet measured.
+- Implication for the gate: v10 must match raw completion and cost per completed task. Any extra model call (planner, Wall author, fix round) adds cost against a $0.19 baseline, so the fast lane must skip stages a small task does not need.
+
+## 2026-09-27T23:27Z: Loki 10 progress
+- Raw claude -p full arm: 27/29 (93.1%), p50 39s, p90 68s, $0.2363 per completed task (docs/v10/METRICS.md).
+- v10 first real task (E-14 glue entry, commit 9d9ff7d2): fixed in 30s, hidden test result "1 passed, exit 0", $0.45 (plan $0.23 plus implement $0.22; `cost-summary.py --json` fully_measured true) against raw $0.13 on the same task.
+- On main (`git ls-remote origin refs/heads/main` = 44731e0f): every engine slice except E-14's glue, plus E-41 local-origin push; v9.79.0 tagged 23:12:55.
+- In flight: E-42 (real entry end to end, Rule of Two for the pr stage, harness v10 arm on one task), E-43 cost knobs, E-44 output fixes, E-33..E-35, EV-5 legacy full arm.
+- ETA: first harness-scored v10 task by about 00:15Z; full v10 arm by about 01:30Z; gate decision 03:00Z.
+- Top blocker: cost. Plan plus implement is about 2x raw per task on this sample; the gate requires v10 cost per completed task at or below raw ($0.2363).
+
+## 2026-09-28T00:11Z: Loki 10 progress
+- On main (`git ls-remote origin refs/heads/main` = addda547): E-42 real entry end to end (run.ts glue deleted), E-44 output fixes, agent SDK 0.3.283 (9db2a4ed), E-51 eval PR path regression, and train E12 (23b20e6f): E-33 gate report, EV-6 publish script, E-38 legacy arm pin, E-39 interrupt/resume, E-40 live PR smoke. Train checks: engine10 349 pass 0 fail, tsc exit 0, test-harness 79/0, test-gate-report 34/0.
+- SDK route: claude-opus-5-5 now runs through the SDK (probe: `success OK`); 0.3.267 rejected it. The same probe cost $0.33 for a one-word reply, so fixed per-call overhead, not task work, drives v10 cost.
+- Legacy arm (EV-5): 3 results so far (2 hidden pass), each at the 900s cap; runner alive at parallel 3, ETA about 02:15Z. Incident: 6 then 2 legacy watchdog loops escaped the harness timeout group (ppid 1, 30 min); stopped by PID; guard slice EV-10.
+- In flight (wave E13, 12 builders): E-45 cost path rework, E-36 preflight (plus E-37), E-32 rebase, E-34 docs refresh, E-52..E-56, E-58 (makes E-47 green), E-59 (makes E-50 green), EV-10.
+- ETA: E-45 merge about 00:45Z, EV-8 5-task cost check right after, EV-9 full v10 arm about 01:00Z to 01:45Z, gate decision 03:00Z.
+- Top blocker: cost per completed task (raw $0.2363); E-45 plus the per-call overhead finding decide it.